AI Ethics Cautiously Assessing Whether Offering AI Biases Hunting Bounties To Catch And Nab Ethically Wicked Fully Autonomous Systems Is Prudent Or Futile

Wanted: AI bias hunters.

That could be a modern-day ad that you’ll begin to see popping up on social media channels and making appearances in various online job postings. This is a relatively new concept or role. It carries with it controversy. Some ardent believers fervently argue that it makes abundant sense and should have been happening all along, while others are rather nervously doing some serious head-scratching and not so sure that this is a good idea.

The gist of the role consists of ferreting out of AI any hidden biases or embedded discriminatory practices. To the rescue come the eager and altogether erstwhile AI biases hunters. They would presumably be computer-savvy bounty hunters. More so, hopefully, steeped in the depths of AI capabilities.

Have gun, will travel, and can via AI-skilled keen marksmanship manage to expose those unsavory and untoward AI biases.

This raises a slew of thorny questions about the sensibility of pursuing such a tactic when it comes to discovering AI biases. As I will discuss momentarily, please know that the advent of AI has also brought with it the emergence of AI biases. A torrent of AI biases. For my ongoing and extensive coverage of AI Ethics and Ethical AI, see the link here and the link here, just to name a few.

How are we to discover that a particular AI system has biases?

You might entrust the AI developers that devised the AI to do so. The thing is, they might be so mired in biases that they themselves cannot recognize the biases within their concocted AI. It all looks good to them. Or they might be so excited about the AI and have a sense of self-pride about it that to have to then take a critical eye to examine it for biases would be difficult and a real downer. Lots of other such reasons might seem to undercut having the AI developers take on this task, including lack of skills to figure out the embedded biases, lack of time in a project to do so, etc.

Okay, so go ahead and hire outside consultants to do the dirty work for you, as it were. Turns out that consultants will happily scrutinize your AI for biases, charging you a pretty penny to do so (lots and lots of pennies). Realize that you need to pay for them to come up to speed with your AI system. You need to then have them rummage around, which might take an untold number of costly laborious hours. Using consultants is an option if you have the budget for it.

Into the potential “gap” of how to find those insidious AI biases come the heroic and dashing AI biases bounty hunters.

You don’t usually pay them upfront. They try to find the AI biases on their own time and have to foot their own bills as they do so. Only if they successfully find biases do they get paid. I suppose you could readily assert that in a suitable manner of thinking, that’s the conventional definition of a bounty hunter. Get paid if you succeed. Don’t get paid if you aren’t successful. Period, end of story.

Bounty programs have existed since at least the time of the Romans and thus we might surmise that they do work, having successfully endured as a practice over all of these years.

Here’s a fascinating piece of historical trivia for you. Reportedly, a posted message during the Roman Empire in the city of Pompeii proclaimed that bounty hunters were needed to find a copper pot that went missing from a small shop. The reward for the recovery of the copper pot was an impressive prize of sixty-five bronze coins. Sorry to say that we don’t know if any bounty hunter found the copper pot and claimed the bronze coins, but we do know that bounty hunting has certainly continued since those ancient times.

In more modern times, you might be aware that in the 1980s there were some notable bounties offered to find computer bugs or errors in off-the-shelf software packages and then in the 1990’s Netscape markedly offered a bounty for finding bugs in their web browser (becoming one of the most prominent firms of that day to do so). Google and Facebook had each opted toward bounty hunting for bugs starting in 2010 and 2013 years, respectively. A few years later, in 2016 even the U.S. Department of Defense (DoD) got into the act by having a “Hack the Pentagon” bounty effort (note that the publicly focused bounty was for bugs found in various DoD-related websites and not in defense mission-critical systems).

Let’s dig more deeply into the bug bounty topic. I realize that I am mainly aiming to talk about AI biases in bounty hunting in this discussion, but there are some quite relevant parallels to the bug bounty arena.

Some are demonstrably puzzled that any firm would want to offer a bounty to find bugs (or, in this case, AI biases) in their systems.

On the surface, this seems like a “you are asking for it” kind of strategy. If you let the world know that you welcome those that might try to find holes in your software, it seems tantamount to telling burglars to go ahead and try to break into your house. Even if you already believe that you’ve got a pretty good burglar alarm system and that no one should be able to get into your secured home, imagine asking and indeed pleading with burglars to all descend upon your place of residence and see if they can crack into it. Oh, the troubles we weave for ourselves.

The same could be said about asking for bounty hunters to find biases in your AI.

First, it perhaps implies that you already believe or even outright know that your AI does have biases. That is a shockingly forthright implied admission that few would seem willing to make and could potentially backfire.

Second, you don’t know for sure what those bounty hunters might do. They could opt to tell the whole world that they found biases in your AI. One supposes this might forfeit getting the bounty, though some might relish the attention or aim to bolster their status for getting consulting gigs and other revenue-generating possibilities. It could perhaps be entirely altruistic. It might be a form of AI activism. I can keep going.

Third, there could be a sneaky twist to the whole affair. A bounty hunter claiming to be searching for AI biases might be devilishly poking around to find ways to attack your AI system. The whole thing is a charade to ultimately undertake a severe cyberattack. You might have assumed they were trying to help, while they have wrongdoing in their hearts. Sad, but possible.

Fourth, we can get even more slyly contorted on this matter. A bounty hunter finds some embarrassing and potentially lawsuit-spurring AI biases. The bounty is some amount of dollars that we will call X. Rather than claiming the bounty, the bounty hunter does a kind of oddish ransomware provocation. If you pay the bounty hunter an amount of ten times X or maybe the skies the limit, they will tell you about the AI biases. You have until Sunday night at midnight to answer. After that point in time, the AI biases will be revealed for all to see. Yikes, a dastardly predicament to be in.

Fifth, the latest are those so-called “hack to return” cybercrooks that after having stolen a bunch of online dough, they decide to repent and return some of the ill-gotten booty that they grabbed up. The company getting its monies partially returned is then willing to consider the remaining stolen amount as an after-the-fact bounty rewarded to the thieves. Seems like everyone “wins” in that the bulk of the funds are given back and meanwhile the cybercrooks are not legally pursued, plus they get the pirate bounty to boot. Is this prudent or insidiously perpetuating wrongful acts?

I realize that some of you might be saying that nobody should be releasing AI that has any biases in it. That would seem to solve this whole dilemma about whether to use AI biases bounty hunters or not. Just don’t get yourself into a bounty situation. Make sure your AI developers do the right thing and do not allow AI biases into their AI systems. Perhaps use consultants to do a double-check. In essence, do whatever you need to do to avoid ever thinking about or asking those AI bias bounty hunters to come to the table.

Yes, that would seem entirely sensible. The problem is that it is also a bit dreamy. The complexity of many AI systems is so great that trying to ensure that not a single ounce of AI biases emerges is going to be arduous to do. On top of that, some AI systems are purposely devised to adjust and “learn” as they go along. This means that at some future point in time the AI that you devised, which let’s pretend at first was purely clean and without biases, might gravitate toward embodying biases (I do not mean that in an anthropomorphic way, as I will explain further as we go along on this topic).

Those that favor bounty hunting for software bugs are prone to argue that it makes sense to offer such bounties. We can consider their rationale and see if it applies to the AI biases realm too.

Proponents of bug bounties emphasize that rather than trying to pretend that there aren’t any holes in your system, why not encourage holes to be found, doing so in a “controlled” manner? In contrast, without such a bounty effort, you could just hope and pray that by random chance no one will find a hole, but if instead you are offering a bounty and telling those that find a hole that they will be rewarded, it offers a chance to then shore-up the hole on your own and then prevents others from secretly finding it at some later point in time.

The same could be said in the use case of AI biases. If you offer a sufficient bounty, hopefully, the bounty hunters will bring the discovery of AI biases to your attention. You can then cope with the AI biases in a relatively quiet and measured way. This might forestall a much larger and more daunting problem later on, namely that somebody else finds AI biases in your AI and screams about it to the high heavens.

Generally, a firm wishing to enable bugs bounty hunting effort will put in place a Vulnerability Disclosure Policy (VDP). The VDP indicates how the bugs are to be found and reported to the firm, along with how the reward or bounty will be provided to the hunter. Usually, the VDP will require that the hunter end up signing a Non-Disclosure Agreement (NDA) such that they won’t reveal to others what they found.

The notion of using an NDA with bounty hunters has some controversy. Though it perhaps makes sense to the company offering the bounty to want to keep mum the exposures found, it also is said to stifle overall awareness about such bugs. Presumably, if software bugs are allowed to be talked about, it would potentially aid the safety of other systems at other firms that would then shore up their exposures. Some bounty hunters won’t sign an NDA, partially due to the public desire and partially due to trying to keep their own identity hidden. Keep in mind too that the NDA aspect doesn’t arise usually until after the hunter claims they have found a bug, rather than requiring it beforehand.

Some VDPs stipulate that the NDA is only for a limited time period, allowing the firm to first find a solution to the apparent hole and then afterward to allow for wider disclosure about it. Once the hole has been plugged, the firm then allows a loosening of the NDA so that the rest of the world can know about the bug. The typical time-to-resolution for bounty hunted bugs is reportedly around 15-20 days when a firm wants to plug it right away, while in other cases it might stretch out to 60-80 days. In terms of paying the bounty hunter, the so-called time-to-pay, after the hole has been verified as actually existing, the bounty payments reportedly tend to be within about 15-20 days for the smaller instances and around 50-60 days for the larger instances (these are constantly changing industry indications and only mentioned as illustrative).

Should AI biases bounty hunters also be asked to participate in a VDP and deal with an NDA?

You can get a yes and a no to that question. Yes, some firms should go that route. No, you might not necessarily opt to go that route. Factors include the size and nature of the AI, the potential of any AI biases exposures involved, and a bunch of other ethical, legal, and business considerations that come to bear.

I might add that establishing a bounty hunting endeavor for AI biases of your AI is a much taller order than you might assume at an initial glance.

We will start with the formidable possibility that you will be overwhelmed by AI biases bounty hunters.

Right now, you would be hard-pressed to find many that would have such a calling card. There aren’t many around. It is the days of the Wild West in that regard. But if the notion of AI biases in bounty hunting catches on, especially when the bounties are plentiful and richly rewarding, you can bet that everyone will dive into the biases hunting swimming pool.

Do you want all sorts of riffraff pursuing AI biases in your AI system? You will get some takers that are actually experts at this sort of thing. You will get other takers that are amateurs and might make a mess or cry wolf. The next thing you know, anybody that can spell “Artificial Intelligence” will be coming to dig in your goldmine of an AI system for those precious AI biases gold nuggets. The gold rush is on. That might not be good for you.

You will need to scrutinize the bounty hunter submissions. There will be a lot of “noise” in the reported claims, in the sense that many of the claimed AI biases do not exist, though the bounty hunter insists that they found some. Imagine how much labor your own AI teams will be required to examine the bounty claims, explore the validity of each, and then potentially go back and forth with the bounty hunter about whether gold was discovered or not.

Some would argue that’s another reason to do the whole thing yourself. You might inevitably discover that the bounty thing is more trouble than it was worth.

Here’s another question to ponder. How will the bounty hunters know what an AI bias looks like? In essence, without some semblance of what to be looking for, any shiny rock could be claimed as showcasing an AI bias in the perceived AI goldmine being excavated.

In the days of the Old West, suppose you offered a reward for the capture of Billy the Kid (a famous outlaw). If you did so and did not include a picture of what Billy looked like, imagine the number of bounty hunters that might drag into the sheriff’s office someone that they hoped or thought was Billy the Kid. You might get inundated with false Billy’s. This is bad since you’d need to presumably look at each one, ask probing questions, and try to ascertain whether the person was really Billy or not.

The point is that to set up the AI biases bounty effort you would be wise to try and clarify what you consider AI biases to consist of. This requires a Goldilocks kind of calibration. You don’t want to be so confining that the bounty hunters overlook AI biases merely because they don’t fit within your stipulated definition, and nor do you want them to yell “Eureka!” at every morsel of an AI bias that they perchance find.

You will need just the right Goldilocks balance of what AI biases consist of and thus provide preferably explicit directions thereof.

A lot of this AI biases bounty hunting is going to be focused on AI-based Machine Learning (ML) and Deep Learning (DL) systems. This makes sense since the advent of ML/DL pervasiveness is growing, plus it seems to have some of the most likely challenges of encompassing undue AI biases.

These researchers identify how salient an AI biases bounty hunting effort can be, particularly in the ML/DL context: “Over time, the software and security communities have developed ‘bug bounties’ in an attempt to turn similar dynamics between system developers and their critics (or hackers) towards more interactive and productive ends. The hope is that by deliberately inviting external parties to find software or hardware bugs in their systems, and often providing monetary incentives for doing so, a healthier and more rapidly responding ecosystem will evolve. It is natural for the ML community to consider a similar ‘bias bounty’ approach to the timely discovery and repair of models and systems with bias or other undesirable behaviors. Rather than finding bugs in software, external parties are invited to find biases — for instance, (demographic or other) subgroups of inputs on which a trained model underperforms — and are rewarded for doing so” (in the paper “An Algorithmic Framework for Bias Bounties” by Ira Globus-Harris, Michael Kearns and Aaron Roth).

In the research paper, the authors outline a suggested approach to what kinds of AI biases can be sought by bounty hunters. There is also an indication about how to assess the bounty hunter claims associated with the alleged AI biases so discovered. As per my earlier remarks herein, the odds are that you will receive specious claims and have to separate the AI biases wheat from the chaff.

Before getting into some more meat and potatoes about the wild and woolly considerations underlying AI bias hunting, let’s establish some additional fundamentals on profoundly integral topics. We need to briefly take a breezy dive into AI Ethics and especially the advent of Machine Learning (ML) and Deep Learning (DL).

You might be vaguely aware that one of the loudest voices these days in the AI field and even outside the field of AI consists of clamoring for a greater semblance of Ethical AI. Let’s take a look at what it means to refer to AI Ethics and Ethical AI. On top of that, we will explore what I mean when I speak of Machine Learning and Deep Learning.

One particular segment or portion of AI Ethics that has been getting a lot of media attention consists of AI that exhibits untoward biases and inequities. You might be aware that when the latest era of AI got underway there was a huge burst of enthusiasm for what some now call AI For Good. Unfortunately, on the heels of that gushing excitement, we began to witness AI For Bad. For example, various AI-based facial recognition systems have been revealed as containing racial biases and gender biases, which I’ve discussed at the link here.

Efforts to fight back against AI For Bad are actively underway. Besides vociferous legal pursuits of reining in the wrongdoing, there is also a substantive push toward embracing AI Ethics to righten the AI vileness. The notion is that we ought to adopt and endorse key Ethical AI principles for the development and fielding of AI doing so to undercut the AI For Bad and simultaneously heralding and promoting the preferable AI For Good.

On a related notion, I am an advocate of trying to use AI as part of the solution to AI woes, fighting fire with fire in that manner of thinking. We might for example embed Ethical AI components into an AI system that will monitor how the rest of the AI is doing things and thus potentially catch in real-time any discriminatory efforts, see my discussion at the link here. We could also have a separate AI system that acts as a type of AI Ethics monitor. The AI system serves as an overseer to track and detect when another AI is going into the unethical abyss (see my analysis of such capabilities at the link here).

In a moment, I’ll share with you some overarching principles underlying AI Ethics. There are lots of these kinds of lists floating around here and there. You could say that there isn’t as yet a singular list of universal appeal and concurrence. That’s the unfortunate news. The good news is that at least there are readily available AI Ethics lists and they tend to be quite similar. All told, this suggests that by a form of reasoned convergence of sorts that we are finding our way toward a general commonality of what AI Ethics consists of.

First, let’s cover briefly some of the overall Ethical AI precepts to illustrate what ought to be a vital consideration for anyone crafting, fielding, or using AI.

For example, as stated by the Vatican in the Rome Call For AI Ethics and as I’ve covered in-depth at the link here, these are their identified six primary AI ethics principles:

  • Transparency: In principle, AI systems must be explainable
  • Inclusion: The needs of all human beings must be taken into consideration so that everyone can benefit, and all individuals can be offered the best possible conditions to express themselves and develop
  • Responsibility: Those who design and deploy the use of AI must proceed with responsibility and transparency
  • Impartiality: Do not create or act according to bias, thus safeguarding fairness and human dignity
  • Reliability: AI systems must be able to work reliably
  • Security and privacy: AI systems must work securely and respect the privacy of users.

As stated by the U.S. Department of Defense (DoD) in their Ethical Principles For The Use Of Artificial Intelligence and as I’ve covered in-depth at the link here, these are their six primary AI ethics principles:

  • Responsible: DoD personnel will exercise appropriate levels of judgment and care while remaining responsible for the development, deployment, and use of AI capabilities.
  • Equitable: The Department will take deliberate steps to minimize unintended bias in AI capabilities.
  • Traceable: The Department’s AI capabilities will be developed and deployed such that relevant personnel possesses an appropriate understanding of the technology, development processes, and operational methods applicable to AI capabilities, including transparent and auditable methodologies, data sources, and design procedure and documentation.
  • Reliable: The Department’s AI capabilities will have explicit, well-defined uses, and the safety, security, and effectiveness of such capabilities will be subject to testing and assurance within those defined uses across their entire lifecycles.
  • Governable: The Department will design and engineer AI capabilities to fulfill their intended functions while possessing the ability to detect and avoid unintended consequences, and the ability to disengage or deactivate deployed systems that demonstrate unintended behavior.

I’ve also discussed various collective analyses of AI ethics principles, including having covered a set devised by researchers that examined and condensed the essence of numerous national and international AI ethics tenets in a paper entitled “The Global Landscape Of AI Ethics Guidelines” (published in Nature), and that my coverage explores at the link here, which led to this keystone list:

  • Transparency
  • Justice & Fairness
  • Non-Maleficence
  • Responsibility
  • Privacy
  • Beneficence
  • Freedom & Autonomy
  • Trust
  • Sustainability
  • Dignity
  • Solidarity

As you might directly guess, trying to pin down the specifics underlying these principles can be extremely hard to do. Even more so, the effort to turn those broad principles into something entirely tangible and detailed enough to be used when crafting AI systems is also a tough nut to crack. It is easy to overall do some handwaving about what AI Ethics precepts are and how they should be generally observed, while it is a much more complicated situation in the AI coding having to be the veritable rubber that meets the road.

The AI Ethics principles are to be utilized by AI developers, along with those that manage AI development efforts, and even those that ultimately field and perform upkeep on AI systems. All stakeholders throughout the entire AI life cycle of development and usage are considered within the scope of abiding by the being-established norms of Ethical AI. This is an important highlight since the usual assumption is that “only coders” or those that program the AI are subject to adhering to the AI Ethics notions. As earlier stated, it takes a village to devise and field AI, and for which the entire village has to be versed in and abide by AI Ethics precepts.

Let’s also make sure we are on the same page about the nature of today’s AI.

There isn’t any AI today that is sentient. We don’t have this. We don’t know if sentient AI will be possible. Nobody can aptly predict whether we will attain sentient AI, nor whether sentient AI will somehow miraculously spontaneously arise in a form of computational cognitive supernova (usually referred to as the singularity, see my coverage at the link here).

The type of AI that I am focusing on consists of the non-sentient AI that we have today. If we wanted to wildly speculate about sentient AI, this discussion could go in a radically different direction. A sentient AI would supposedly be of human quality. You would need to consider that the sentient AI is the cognitive equivalent of a human. More so, since some speculate we might have super-intelligent AI, it is conceivable that such AI could end up being smarter than humans (for my exploration of super-intelligent AI as a possibility, see the coverage here).

Let’s keep things more down to earth and consider today’s computational non-sentient AI.

Realize that today’s AI is not able to “think” in any fashion on par with human thinking. When you interact with Alexa or Siri, the conversational capacities might seem akin to human capacities, but the reality is that it is computational and lacks human cognition. The latest era of AI has made extensive use of Machine Learning (ML) and Deep Learning (DL), which leverage computational pattern matching. This has led to AI systems that have the appearance of human-like proclivities. Meanwhile, there isn’t any AI today that has a semblance of common sense and nor has any of the cognitive wonderment of robust human thinking.

ML/DL is a form of computational pattern matching. The usual approach is that you assemble data about a decision-making task. You feed the data into the ML/DL computer models. Those models seek to find mathematical patterns. After finding such patterns, if so found, the AI system then will use those patterns when encountering new data. Upon the presentation of new data, the patterns based on the “old” or historical data are applied to render a current decision.

I think you can guess where this is heading. If humans that have been making the patterned upon decisions have been incorporating untoward biases, the odds are that the data reflects this in subtle but significant ways. Machine Learning or Deep Learning computational pattern matching will simply try to mathematically mimic the data accordingly. There is no semblance of common sense or other sentient aspects of AI-crafted modeling per se.

Furthermore, the AI developers might not realize what is going on either. The arcane mathematics in the ML/DL might make it difficult to ferret out the now hidden biases. You would rightfully hope and expect that the AI developers would test for the potentially buried biases, though this is trickier than it might seem. A solid chance exists that even with relatively extensive testing that there will be biases still embedded within the pattern matching models of the ML/DL.

You could somewhat use the famous or infamous adage of garbage-in garbage-out. The thing is, this is more akin to biases-in that insidiously get infused as biases submerged within the AI. The algorithm decision-making (ADM) of AI axiomatically becomes laden with inequities.

Not good.

Let’s now return to the topic of AI bias hunting.

For those of you considering an AI bias bounty hunting endeavor, here are my recommended seven key steps on how to best proceed:

1) Assess. Assess the suitability of an AI bias bounty hunting endeavor for your circumstances and as per your AI systems

2) Design. Design an appropriate AI bias bounty hunting approach

3) Implement. Implement and publicize your AI bias bounty hunting endeavors

4) Field. Field the AI biases bounty claims and process accordingly

5) Fix. Fix or adjust your AI as pertinent to these discovered AI bias exposures

6) Adjust. Adjust the AI biases bounty hunting as needed

7) Discontinue. Discontinue the AI bias bounty hunting when it is no longer needed

In my series of above steps, note that I mention that you will presumably want to fix or adjust your AI as based on ascertaining that a claimed AI bias does in fact exist within your AI system. This abundantly makes sense. You would almost certainly want to shore up any found AI biases. Think of the legal (and ethical) ramifications if you don’t do so. It is one thing to assert that you didn’t know an AI bias existed and therefore allowed it to exist, while it is much shakier ground to have on record that you were made aware of an AI bias and did nothing about it.

The nature and degree of the AI fix or adjustment would of course be contingent on how significant the AI biases were and how deeply embedded the issues are. If you are lucky, perhaps a modest amount of changes to the AI will rectify matters. The other potential is that you might need to do an entire rewrite of the AI. For the ML/DL type of AI, this could require going back to the drawing board and starting fresh with an entirely new set of data and a cleaned-up ML/DL model. I’ve discussed the advent of AI disgorgement or AI destruction as a potential legal remedy against unsavory AI, see the link here.

One question to mull over is whether you would want the bounty hunters to possibly do more than just identify the existence of AI biases. For example, you might sweeten the bounty by indicating that proposed fixes are welcome too. An AI bias found by a bounty hunter might be paid one indicated reward or prize. If the bounty hunter can also proffer a viable fix to the AI bias they might then be granted an additional reward.

Some argue that this is a bridge too far. They say that you should keep the AI bias bounty hunters exclusively focused on finding AI biases. You are going to create a bunch of undesirable adverse consequences by inviting them to also suggest fixes. Keep things simple. The goal is to get as many additional eyes on discovering AI biases so that you can decide what to do next. Do not muddy the waters.

A thorny aspect that needs to be figured out entails the magnitude of the reward or prize for the bounty hunters that genuinely discover AI biases. You want the payoff to be demonstrative. Without a high enough reward, you won’t get many bounty hunters or they will not be especially eager to seek out the AI biases in your AI systems. They might instead concentrate on other AI bias bounty endeavors.

Furthermore, as mentioned, you want to try and repress an urge by the bounty hunters to turn their AI bias discoveries into other forms of gold. If the reward seems measly, it could irk bounty hunters into seeking other higher payoffs. They could take a ransomware approach toward you. They might declare that they have a juicy AI bias that a competitor would love to know about and could use against your firm by touting that the AI bias exists in your AI. Thus, they sell the discovered AI bias to the highest bidder. And so on.

One supposes that if you set the reward at an extremely high range, you are also asking for potential trouble. This could attract all kinds of nutty bounty hunters. They in turn might deluge social media with hazy claims that they found a multitude of AI biases, doing so for their own self-promotion and without having actually speared any AI biases. In a sense, your heightened reward inadvertently shines a light on your AI and prods a slew of uncouth moths to correspondingly be perniciously attracted to the glowing light beam.

Another consideration involves accessibility to your AI.

To enable an AI bounty hunting possibility, the bounty hunters have to sufficiently gain access to your AI. They aren’t going to have much luck in finding AI biases if they are entirely locked out. But you don’t want to give up your cybersecurity protections since doing so could completely compromise your AI system.

You might try to have the bounty hunters sign various legally binding declarations and then provide them with the needed access. Some bounty hunters aren’t going to like that type of approach. Their viewpoint is that they will only do whatever any publicly available and open-ended path allows. They are free mavericks, as it were, and do not like being saddled, as it were. Getting them to put their signature on intimidating legal documents will cause a lot of them to avoid searching for AI biases in your AI. Or they might get peeved at your legal gauntlet and decide they will see what they can find via public means, doing so with the perhaps strident urge to show you how vulnerable you really are.

I’ve got yet another angle that might make your head spin.

An AI savvy bounty hunter might decide to devise an AI system that can scrutinize your AI and possibly discover AI biases in your AI. This is the toolmaker opting to make a tool to do the job rather than performing manual labor themselves. Instead of laboriously examining your AI, the AI-versed bounty hunter spends their time concocting an AI tool that does the same thing. They then use the AI tool on your AI. The beauty too is that they can presumably reuse the AI tool on anyone else that is also offering a bounty hunting opportunity on their respective AI as well.

I know what you are probably thinking. If an AI tool can be devised to examine AI for biases, the maker of the AI that is being scrutinized for AI biases ought to either craft such an AI tool or buy one for their own use. In theory, they then do not need to contend with the whole bounty hunter carnival, to begin with. Just use AI to find their AI biases.

Yes, this is something that you can expect will gradually arise. Meanwhile, the mainstay of these efforts will likely consist of AI developers doing bounty hunting. They might use various tools to aid their efforts, but in the near term, they are unlikely to simply mindlessly set the AI tool on automatic and take a nap such that the tool does the entirety of the AI bias hunting for them.

We aren’t there yet.

At this juncture of this weighty discussion, I’d bet that you are desirous of some illustrative examples that might showcase this topic. There is a special and assuredly popular set of examples that are close to my heart. You see, in my capacity as an expert on AI including the ethical and legal ramifications, I am frequently asked to identify realistic examples that showcase AI Ethics dilemmas so that the somewhat theoretical nature of the topic can be more readily grasped. One of the most evocative areas that vividly presents this ethical AI quandary is the advent of AI-based true self-driving cars. This will serve as a handy use case or exemplar for ample discussion on the topic.

Here’s then a noteworthy question that is worth contemplating: Does the advent of AI-based true self-driving cars illuminate anything about the use of AI bias bounty hunting, and if so, what does this showcase?

Allow me a moment to unpack the question.

First, note that there isn’t a human driver involved in a true self-driving car. Keep in mind that true self-driving cars are driven via an AI driving system. There isn’t a need for a human driver at the wheel, nor is there a provision for a human to drive the vehicle. For my extensive and ongoing coverage of Autonomous Vehicles (AVs) and especially self-driving cars, see the link here.

I’d like to further clarify what is meant when I refer to true self-driving cars.

Understanding The Levels Of Self-Driving Cars

As a clarification, true self-driving cars are ones where the AI drives the car entirely on its own and there isn’t any human assistance during the driving task.

These driverless vehicles are considered Level 4 and Level 5 (see my explanation at this link here), while a car that requires a human driver to co-share the driving effort is usually considered at Level 2 or Level 3. The cars that co-share the driving task are described as being semi-autonomous, and typically contain a variety of automated add-ons that are referred to as ADAADA
S (Advanced Driver-Assistance Systems).

There is not yet a true self-driving car at Level 5, and we don’t yet even know if this will be possible to achieve, nor how long it will take to get there.

Meanwhile, the Level 4 efforts are gradually trying to get some traction by undergoing very narrow and selective public roadway trials, though there is controversy over whether this testing should be allowed per se (we are all life-or-death guinea pigs in an experiment taking place on our highways and byways, some contend, see my coverage at this link here).

Since semi-autonomous cars require a human driver, the adoption of those types of cars won’t be markedly different than driving conventional vehicles, so there’s not much new per se to cover about them on this topic (though, as you’ll see in a moment, the points next made are generally applicable).

For semi-autonomous cars, it is important that the public needs to be forewarned about a disturbing aspect that’s been arising lately, namely that despite those human drivers that keep posting videos of themselves falling asleep at the wheel of a Level 2 or Level 3 car, we all need to avoid being misled into believing that the driver can take away their attention from the driving task while driving a semi-autonomous car.

You are the responsible party for the driving actions of the vehicle, regardless of how much automation might be tossed into a Level 2 or Level 3.

Self-Driving Cars And AI Bias Bounty Hunting

For Level 4 and Level 5 true self-driving vehicles, there won’t be a human driver involved in the driving task.

All occupants will be passengers.

The AI is doing the driving.

One aspect to immediately discuss entails the fact that the AI involved in today’s AI driving systems is not sentient. In other words, the AI is altogether a collective of computer-based programming and algorithms, and most assuredly not able to reason in the same manner that humans can.

Why is this added emphasis about the AI not being sentient?

Because I want to underscore that when discussing the role of the AI driving system, I am not ascribing human qualities to the AI. Please be aware that there is an ongoing and dangerous tendency these days to anthropomorphize AI. In essence, people are assigning human-like sentience to today’s AI, despite the undeniable and inarguable fact that no such AI exists as yet.

With that clarification, you can envision that the AI driving system won’t natively somehow “know” about the facets of driving. Driving and all that it entails will need to be programmed as part of the hardware and software of the self-driving car.

Let’s dive into the myriad of aspects that come to play on this topic.

First, it is important to realize that not all AI self-driving cars are the same. Each automaker and self-driving tech firm is taking its approach to devising self-driving cars. As such, it is difficult to make sweeping statements about what AI driving systems will do or not do.

Furthermore, whenever stating that an AI driving system doesn’t do some particular thing, this can, later on, be overtaken by developers that in fact program the computer to do that very thing. Step by step, AI driving systems are being gradually improved and extended. An existing limitation today might no longer exist in a future iteration or version of the system.

I hope that provides a sufficient litany of caveats to underlie what I am about to relate.

In my columns, I have already discussed at length the use of bugs-oriented bounty hunters in the autonomous vehicles and self-driving cars realm. This approach has indeed taken place in this niche. There are the usual debates about whether it is a sound idea or not. The efforts have usually been of a limited nature, often kept relatively quiet.

A likewise discourse can ensue when the focus shifts toward hunting for AI biases rather than seeking system bugs per se. Some suggest it is a darned if you do, darned if you don’t conundrum.

Here’s why.

First, to be clear, there are numerous ways in which autonomous vehicles and self-driving cars are going to be subject to containing AI biases, see my coverage at the link here and the link here, just to name a few. Automakers and self-driving car firms would seem wise to try and prevent those AI biases from appearing in their AI systems. The legal and ethical firestorm against such firms will undoubtedly be intense.

Is the use of an AI biases bounty hunting endeavor a suitable approach in this specific context?

One answer is that yes, this will be handy and provide an abundance of “free” sets of new eyes to try and catch any embedded AI biases of an AI self-driving car or the like. Most of the AI developers building self-driving cars are busy making AI that can safely drive a car from point A to point B. They are preoccupied with that core capability and have neither the time nor attention toward any AI biases that might be somewhere in their AI.

The other answer is that no, allowing bounty hunting for autonomous vehicles and self-driving cars on any basis, whether for bugs or AI biases, just ought to be fiercely avoided. The argument is that these vehicles and their AI are of a life-or-death caliber. Messing with the AI in any fashion could be somehow ruinous to the AI and impact what the AI driving system does.

A counterargument to that last point is that the bounty hunters are supposed to be unable to alter the AI that they are examining. Thus, there is no danger of them messing with the AI and causing the AI in this context to suddenly become a crazed AI driving system. The bounty hunters are only having read-only access. Allowing them to go further would be amply stupid and a huge mistake.

The counterargument to that counterargument is that by allowing and encouraging bounty hunters to examine your AI, the whole matter becomes dicey. Those bounty hunters might figure out ways to exploit any found bugs or biases. Those exploits in turn might be for devious purposes. You would be better off not inviting “burglars” into your home, so to speak. Once they have cased out the joint, you are ultimately going to be in a heap of trouble.

For those that have AI systems of a less-than-life-or-death magnitude, the belief is that the repercussions of a bounty hunting foray that goes awry are a lot less risky. Maybe so. On the other hand, if a firm has poured their monies into an AI system that bounty hunters manage to usurp, you can assume that the reputational damages and other potential damages will still hurt.

There is no free lunch when it comes to AI bias bounty hunting.

A quick closing remark for now.

When the notorious outlaw Jesse James was sought during the Old West, a “Wanted” poster was printed that offered a bounty of $5,000 for his capture (stating “dead or alive”). It was a rather massive sum of money at the time. One of his own gang members opted to shoot Jesse dead and collect the reward. I suppose that shows how effective a bounty can be.

Will the use of AI bias bounty hunters be a good thing, or will it be a bad thing?

If you opt to institute an AI bias bounty hunter endeavor, I’d suggest that you keep your eyes wide open and be looking over your shoulder at all times. This is prudent for you and your AI. You never know what might happen, including that a conniving bounty hunter somehow inserts surreptitiously an AI bias into your AI and shouts to the world that they found an unscrupulous AI bias in your AI. Perhaps doing so in a brazen and outsized attempt at seeking the bounty reward, plus proclaiming themselves a hero that essentially got the vaunted Jesse James.

Come to think of it, a sentient AI probably won’t like that idea of a disconcerting dead-or-alive provision, one might languorously so speculate.

Source: https://www.forbes.com/sites/lanceeliot/2022/07/16/ai-ethics-cautiously-assessing-whether-offering-ai-biases-hunting-bounties-to-catch-and-nab-ethically-wicked-fully-autonomous-systems-is-prudent-or-futile/