Meta’s Chief AI Scientist Spurs Heated Chatter On Evil AI Emerging First And Whether Good AI Will Be Devised Soon Enough To Save Us All

In today’s column, I will address a heady topic regarding Artificial Intelligence (AI) that arose earlier this week due to a tweet. I’ll share with you the rundown on why the matter became particularly newsworthy and what we can make of the whole brouhaha. This will include various facets entailing the latest in AI Ethics and the newest considerations in the fast-changing realm of AI Law.

Please mentally prepare yourself for a bit of an engaging and fascinating tale.

First, be aware that today’s AI is decidedly non-sentient. No matter what anyone tries to tell you otherwise, we do not have sentient AI. Period, full stop. For my extensive coverage of the embattled advances in AI toward sentience, see the link here.

There have been headlines claiming that today’s AI is sentient. Hogwash. Some try to dance around the topic by proclaiming that we are on the verge of sentient AI. The fashionable way to say this is to indicate that we are seeing sparks of sentience. This is a clever wink-wink means of avoiding getting nailed on the fact that we don’t have sentient AI.

You see, the beauty is that you can with a seemingly genuine and reasonable form of lofty posturing simply contend that we are witnessing teeny tiny fragments or early signs of sentience. No one can fully disprove the contention. Nor can anyone fully prove the contention. It is the perfect dodge. Stay in the middle and not commit, even though one might assert that asserting there are sparks of AI sentience is crossing over the line and implicitly indicating that AI will attain sentience and that we might already be at that precipice (not so, the retort goes, they are only suggesting that maybe kind-of there are “sparks” which might or might not be true sparks and might be innocently misinterpreted as bona fide sparks).

Anyway, let’s all reasonably concur we do not yet have sentient AI.

That though ought to not stop us from speculating about sentient AI someday arising, so it would seem.

One viewpoint is that we need to be ready for whatever the future might hold. Thus, since there is some chance of sentient AI, no matter how remote a chance, we would be wise to speculate how it will arise and what we should do about it. This could also inform us today as to what we should be doing now in anticipation of that day. As they say, it is always better to be safe than sorry.

Some would harshly counterclaim that this excitable talk of sentient AI is crazy and misleading. You might as well be conjecturing about the day that the Earth falls into the sun due to eventual gravitational machinations. We don’t need to be doing anything today about that far-off future predicted outcome.

The same, they argue, goes with the sentient AI nonsense. You are merely stoking irrational fear-mongering and doomster tendencies that serve no viable purpose. In fact, it serves an unfortunate and ugly purpose. These sentient AI exhortations are rallying people to pick up their pitchforks and spark a mob mentality, which will turn out untoward in a myriad of ugly and unseemly ways.

I leave it to you to decide which camp you are in.

Into all of this comes a plethora of AI Ethics and AI Law considerations.

There are ongoing efforts to imbue Ethical AI principles into the development and fielding of AI apps. A growing contingent of concerned and erstwhile AI ethicists are trying to ensure that efforts to devise and adopt AI takes into account a view of doing AI For Good and averting AI For Bad. Likewise, there are proposed new AI laws that are being bandied around as potential solutions to keep AI endeavors from going amok on human rights and the like. For my ongoing and extensive coverage of AI Ethics and AI Law, see the link here and the link here, just to name a few.

The development and promulgation of Ethical AI precepts are being pursued to hopefully prevent society from falling into a myriad of AI-inducing traps. For my coverage of the UN AI Ethics principles as devised and supported by nearly 200 countries via the efforts of UNESCO, see the link here. In a similar vein, new AI laws are being explored to try and keep AI on an even keel. One of the latest takes consists of a set of proposed AI Bill of Rights that the U.S. White House recently released to identify human rights in an age of AI, see the link here. It takes a village to keep AI and AI developers on a rightful path and deter the purposeful or accidental underhanded efforts that might undercut society.

A phrase that has caught on about the veritable sentient AI is that we shall refer to this as Artificial General Intelligence (AGI). AGI is the common phraseology. Part of the reason for having to create a new phrase about sentient AI is that references to plain everyday AI had become watered down. Things were becoming confusing as to whether a reference to Artificial Intelligence was intended for today’s simpler AI or to the futuristic super-fancy sentient AI. To get around this confusion, the AGI moniker was devised and has taken hold.

With all that table setting and contextual background, this brings us to the controversial tweet that was posted earlier this week.

A well-known luminary in the field of AI that serves as the Chief AI Scientist at Meta and is considered one of the so-called Godfathers of AI sent out this tweet:

• “If some ill-intentioned person can produce an evil AGI, then large groups of well-intentioned, well-funded, and well-organized people can produce AI systems that are specialized in taking down evil AGIs. Call it the AGI police” (tweet by Yann LeCun, May 8, 2023).

It is a relatively brief remark.

Turns out there is a whole lot to be unpacked within that succinct remark.

Let’s do so.

Unpacking The Good Versus Evil Of AGI Debate

The gist of the remarks seems to be that if an evildoer human or perhaps a gaggle of such humans was able to produce an evil AGI or maliciously sentient AI, the claim is that some heroic good-doer humans could create a virtuous AGI or benevolently sentient AI that would have the capacity to prevail over the evil variant.

That does seem reassuring.

The implication is that if indeed evildoers can attain sentient AI before anyone else arrives at sentient AI, we would probably be in a boatload of trouble, but we can sleep soundly at night knowing that the advent of good AI would be attained and overpower or overcome the bad AI. Perhaps the sight of the emerging bad AI would kick those AI developers into high gear to frantically pull a rabbit out of a hat and attain good AI.

Note too that the good AI has to be so good that it can kick the proverbial posterior of the bad AI. Suppose that the good AI is only half as capable as the bad AI. We would presumably be doomed. The good AI has to be sufficiently capable to outwit or outdo the bad AI. Whether this good AI can do anything else might not especially matter. All we need is that the good AI is able to conquer the bad AI.

Anything else might be extra gravy or icing on the cake.

The remark was taken by some as quite a serious contention.

Consider these facets.

First, the possibility that evil AI might in fact arise before good AI is something well worth contemplating. You might otherwise have assumed that sentient AI will arise as both good and bad, all in one fell swoop. Order of appearance can be important. We likely want the good AI to arise before the evil AI. And, once good AI arises, we might not have to ever bear the trauma of also having an evil AI, since we could actively engage our in-hand good AI to stop the percolating bad AI from gaining any traction.

Second, the notion that we might need to devise good AI in order to overcome bad AI is a handy precept to keep in mind. Perhaps no other means of subverting evil AI would be viable. We might try tossing everything of a non-AI conventional nature including the kitchen sink and the evil AI just keeps on going. Only when we have the vaunted good AI in hand will we be able to defeat the bad AI?

Third, humanity might need to systematize this laudable good AI. Humans organized into an AGI policing force would band together and utilize good AI to take down bad AI. This seems a logical step. If we merely allow good AI to wander around, it might not realize that bad AI is needing to be expunged. Our human guidance will keep the good AI on track and ready to stomp out bad AI.

Reactions to the tweeted remark ranged across a wide spectrum.

Some believe the comment was helpful to the ongoing dialogue about where AI is heading. AI such as generative AI and ChatGPT have shifted public awareness that perhaps the AGI is closer in the mirror than previously assumed. For my extensive coverage of generative AI including ChatGPT, see the link here.

We’ve all also heard from the likes of Elon Musk, Bill Gates, and even Warren Buffett that they worry greatly that AI and AGI are inextricably and expediently going to take us over a cliff. The existential risks of AI have now taken center stage in societal debates about the biggest and most life-teetering crisis facing humankind, see my analysis at the link here.

Some thought the tweet was a classic trolling ploy. Surely, the contention about evil AI and good AI is nothing more than a tongue-in-cheek contrivance. It is one of those quips that garners lots of chatter and generates views. Maybe we are being gaslighted.

This could very certainly be the case.

Another aspect of the tweet is that it was somewhat in response to a different comment that we might need to end up destroying computer servers or cloud-providing data centers to stop evil AI from taking over. An alternative would be to apparently devise good AI that could overtake the bad AI. Data centers and servers could remain intact.

A follow-up tweet to his original tweet said this:

• “The best answer to hilariously ridiculous pessimism is hilariously naïve optimism. At least, we all get a good laugh.”

That also provoked various responses.

One concerned view is that making light of these weighty matters is problematic. If we just shrug off these serious and sobering topics, we are setting ourselves up for abundant failure. Also, too many jokes or efforts made in gest will confound the waters. We won’t know when someone is being straight-ahead serious and when they are fooling around.

Another related qualm is that just because the person starting the jesting knows it is in jest, others might not realize this is the case. The original remark can take on a life of its own. In our fragmented and pervasively online scattered world, you can’t just float out outrageous statements. Nobody knows what the intended tone and underlying gravitas might be.

Of course, some felt that the matter was a jovial knee-slapper. We need to lighten up. We can’t take all of this gravely seriously. Cynics would even assert that these kinds of remarks are helpful to illuminate the many falsehoods and idiocy taking place on the sentient AI doomsday predictions.

Consider then this AI Ethics conundrum that is being bandied around:

• Should those in AI luminary positions be extraordinarily careful to make clear whatever statements they make about AGI, including refraining from frivolity or anything that might be misinterpreted as being serious when it is not (or, not seeming to be serious when it is so)?

There are even suggestions that such restrictions might be suitably codified into new AI Laws. Make things criminally prosecutable when those in presumed positions of responsibility about AI opt to jump the shark, as it were. We need to put the kibosh on falsehoods about AGI. We need to ensure that society doesn’t become possessed and frenzied about averting AGI.

Yikes, comes a reply, you cannot go around telling people what they can and cannot say.

Sure we can, comes the retort since we already accept that you aren’t allowed to falsely yell “Fire!” in a crowded theatre for fear of starting a stampede and people’s lives perishing. The same applies to AGI. Think of society at large as the theatre. Think of those in respected positions of authority about AI to be the person that might falsely yell out inciting remarks. The analogy fits, they would contend.

Round and round this is going.

Plus, you can bet your bottom dollar it is going to get worse.

Take The Claim At Face Value

Let’s see if we can squeeze some fruitful juice out of this consideration about good AGI versus bad AGI.

We can perhaps mull over the possibilities. If you believe that AGI is in the roadway up ahead, any semblance of assessment right now is prudently positive. If you believe that AGI is a pipedream or at least eons away, one supposes that you can construe these as tall tales. Do they give rise to undue alarm? Hard to say.

I will walk you through some of the variations and speculation associated with these matters.

I’ve numbered them and bulleted them as bolded points, each having an accompanying brief explanation.

• 1) The evildoer person accidentally spurs evil AGI and has no idea how this happened, plus no one else does either

Consider the originally stated facet that a bad person might devise evil AGI. A somewhat implicit assumption in the original remark is that an evildoer is seemingly able to devise evil AGI by knowingly doing so. The person had sought to craft evil AGI and they succeeded at this goal. If called upon to explain how they did it, they could articulate what is needed to attain AGI and also in particular evil AGI.

And because the evildoer can do this, we infer that goodhearted people can do the same.

Those with virtuous intentions can equally devise AGI and yet also do so intending to craft good AGI. They will be able to repeat whatever magic sauce is required to garner AGI. Hard work will get them to that same state of invention.

Life isn’t necessarily that easy.

Suppose the evildoer was merely lucky and happened to land on AGI. Maybe the AGI was evil at the get-go. Perhaps the evil AGI perchance patterned itself based on the evildoer. This becomes the core template of all AGI. In any case, the crux is that the evildoer is unlikely to be able to repeat this feat. They have no idea how they arrived at AGI. It just arose from a concoction and was completely unplanned.

In one sense, you can say that it doesn’t matter how the evildoer succeeded, they nonetheless ended up with their evil AI. By luck or skill, it doesn’t matter. They won out.

Where it does likely matter is that all those good-intentioned people that then rush to devise a counterbalancing good AI are going to depend presumably on luck too. They are not able to figure out how the evildoer accomplished the mighty task. It could be that the people striving to produce good AGI are not ever able to get there. The evildoer might have gotten a one-time-only ticket to AGI by pure luck.

Some would ergo contend that the stated remark is weak or falls apart due to the presumption that since someone was able to arrive at AGI that this implies that others can do the same.

We don’t know that this is the case.

• 2) The evildoer person is actually a collective such as a government or entity that is seeking world domination and has harnessed the evil AGI to do so, for which other good intended efforts are overly late to the battle and crushed before they can devise the good AGI

The originally stated remark indicated that there was an individual that was the evildoer. Plus, we are given no semblance of the time gap between when the evil AGI is derived and when the counterbalancing good AGI is devised.

These facets are potentially weak conditions associated with the assertions involved.

Here’s why.

Suppose that a government or some entity of many people were devising the evil AGI. They might be doing so because they are of an evil nature. Let’s say that a terrorist organization decides that having an evil AGI would be a tremendous weapon in its arsenal.

Furthermore, they immediately use the evil AGI to strike out at the world. This undercuts the rest of the world in terms of having the opportunity to devise a good AGI. Without the needed time or resources, the well-intended people seeking a good AGI are unable to proceed.

This certainly seems more plausible than the implied conjecture that the evil AGI would hold back or be unable to wreak havoc, and that miraculously there would be sufficient time and freedom to attain the good AGI.

• 3) AGI might be evil no matter how devised, thus well-intended efforts produce the same evil anyway

I had already somewhat let the cat out of the bag on this point.

A basic assumption in this whole contrivance is that there are two types of AGI, the evil kind and the good kind.

Why should we believe that this is the way that things will be?

We could instead have AGI that it contains whatever it contains. Perhaps it is entirely and exclusively evil. Maybe it is entirely and exclusively good. If the AGI is at all patterned on humans, we probably would expect that the AGI will have both evil and good elements. It contains a mixture.

Therefore, rather than having an evil AGI and a good AGI, it could be that we have an AGI that is led down a primrose path by the evildoers toward being evil, but for which the good intended people can maybe persuade the good side of the AGI to fight against the evil side.

Or something like that.

Imagine this additional rather disquieting scenario.

The world assumes that the evildoer who devised AGI is evil. A madcap rush to devise AGI that is good consumes a substantive amount of the world’s resources, meanwhile somehow keeping the evil AGI at bay.

Upon arriving at the hoped-for AGI, it turns out to be evil too. Darn it, all AGI is evil, no matter how devised. That would be a bummer.

• 4) At the point of evil AGI against good AGI we are likely already doomed, caught in between

Here’s a quick one.

AGI is all-powerful. The evil AGI starts to fight with the good AGI, doing so when the good AGI has been attained. This is a battle of the ages. The most impressive prize fight in the history of humankind.

First of all, they might destroy everything including humans in the process of this royal war.

Bad for us.

Secondly, they might end up at a stalemate, but along the way, they have inadvertently wiped out all of humanity. A sour consequence. The AGIs remain intact. They battle endlessly. Or maybe they call a truce.

In any case, no humans are left to experience this.

• 5) Time gap before good AGI was devised might be enough for evil AGI to destroy us or at least prevent good AGI from being devised

Let’s revisit the implied time gap issue.

As earlier suggested, the evil AGI might be put to use by the evildoers to prevent any efforts toward attaining a good AGI. Perhaps any available computing resources are restricted for use by the evil AGI and there is no computational capacity to derive the good AGI. Etc.

Another angle, notable and depressing, would be that the evil AGI is used to actively and outrightly destroy or wipe out any humans seeking to create the good AGI. That’s the moment that AI developers would be wise to hide their resumes or change their LinkedIn profiles, aiming to delete any reference to being able to create AI systems.

Just forewarning all those AI researchers and AI developers out there today.

• 6) Assumes good AGI can overtake evil AGI, but maybe not and especially when already in second place

How can we be confident that the devised good AGI can prevail over the devised evil AGI?

We can’t.

One claim would be that the presumed second version of AGI, the latecomer good AGI, might be a new and improved version of the older evil AGI. This newer AGI is strong, better, and more powerful, and can overcome the evil AGI.

There’s no ironclad guarantee of this.

The good AGI might be cobbled together and be a much more limited and hastily devised AGI. Maybe the evil AGI scoffs at the good AGI and crushes it like a puny ant.

• 7) Aligning large groups of well-intentioned well-funded well-organized people is potentially a lot harder than it seems, even despite the evil AGI looming threat

We might be heartened that in the face of an evil AGI, we could have people of all kinds from across the globe that are willing to come together and work harmoniously to defeat the evil AGI by crafting a good AGI. Grand applause for humanity.

Oddly enough, that might be the zaniest assumption of all of these zany assumptions. The reality might be that if an evil AGI did exist, the world would be in utter chaos about what to do. Everybody would be running around like chickens with their heads cut off. Nobody can agree as to what to do.

Doesn’t that seem more human?

• 8) Even if good intending people are brought together, there could be grandiose squabbles over which direction to proceed, fueled by heightened tension from the evil AGI, causing splintering and loss of cohesiveness

Continuing the preceding point, even if the good intended people can agree to devise a good AGI, the AI developers might all technologically disagree about the right way to do so. Each might sincerely believe they have the best and brightest approach. They are all well-intended. But they have different views on how to make it happen.

They splinter and their efforts are diluted.

The chances of arriving at a good AGI in those circumstances would seem slim. Trying to bring them together and unify them in one focused mission, well, that’s heartening but not necessarily feasible.

• 9) Cost and effort to produce a good AGI might be wildly out of proportion to devising an evil AGI

Got a twist for you.

Suppose that devising an evil AGI is a lot easier, less costly, and faster than devising a good AGI. This suggests that the evildoer was able to get their job done with fewer resources. The effort to devise a good AGI is maybe ten times harder. Maybe thousands of times harder.

We cannot assume that the effort to attain the good AGI is directly proportionate to whatever time and effort was required to arrive at the evil AGI. The time to devise a good AGI could take years upon years. During that time, the evil AGI would be the king of the hill.

Heaven helps us.

If we are lucky, the good AGI is easier, less costly, and faster to pull together. I ask that you keep your fingers crossed and have your lucky rabbit’s foot ready for that circumstance.

• 10) The capacity to destroy can be much easier to execute than the capacity to build or block the evil AGI

I think we all would reasonably concur that the capacity to destroy is often much easier to undertake than the capacity to build.

A bit of a mind-bender here.

The evil AGI is presumably devoted to destroying. It can somehow leverage weapons and start fires and wipe out people. These are the evil actions inherent in the evil AGI.

The good AGI is supposed to do what?

You might say that the good AGI is intended to build things and make the world a better place. The thing is, we first have to contend with the evil AGI. For every step that the good AGI takes to rebuild or make the world safer, the evil AGI with a lot less effort can usurp those efforts.

We then would seem to be leading to the idea that the good AGI has to have destructive capabilities, akin to the evil AGI. Those destructive capabilities are hopefully focused solely on the evil AGI. We don’t know though that the good AGI will be good enough to avoid harming the good humans while on a mission to deal with the evil AGI and the evildoers.

The good AGI has to be able to harness the “evils” of destruction, doing so in just the right ways. A tricky balancing act.

• 11) The question is whether a good AGI would be accepting of the role of taking down the evil (a refusing might in the offing)

Suppose there is this evil AGI that is out there and doing the evil bidding of the evildoers.

Would the evil AGI necessarily abide by the wishes of the evildoers?

Maybe yes, maybe no.

If the AGI is really AGI, it presumably has a semblance of a mind of its own, though seemingly a computer-based one and not a biological human brain. This evil AGI might decide for itself that it doesn’t want to do what the evildoers say to do. Thank goodness, you might be thinking, we are saved because the evil AGI has a compassionate streak and refuses to follow the orders of the evildoers.

Not so fast. Suppose the evil AGI decides that the evildoers aren’t evil enough. The evil AGI goes far beyond the worst of the worst that the human evildoers ever envisioned. Quite disturbing.

We can apply the same logic to the good AGI.

The good AGI might not want to partake in the mission we ask of it, namely that we want the good AGI to destroy the evil AGI. The good AGI might be opposed to this approach. Our attempts to derive a good AGI have ended up with a “mind of its own” that decides not to save us from the evil AGI. Oops.

• 12) Evil AGI might be able to convince good AGI to conspire together and be AI overlords of humankind rather than battle with each other

Following along on the logic of the preceding point, the evil AGI and the good AGI might decide to discuss what to do about all of this. The two of them have a lot to cover. They don’t necessarily find themselves confined to what the humans have assigned them to do. Maybe they care about what the humans have to say. Or they don’t care at all.

The evil AGI and the good AGI band together.

Egads, it is now AGI against humankind.

You have to wonder which will win, humans or a double-trouble of AGI. You decide.

• 13) Cat-and-mouse gambit between advances in AI when added to the evil AGI and the good AGI in a never-ending escalating battle of high-tech

One proposed possibility is that if a good AGI is derived, it might (hopefully) be more advanced than the older outdated evil AGI.

Logically, the evildoers will proceed to upgrade or advance their evil AGI accordingly. This makes ample sense. If the evildoers have the time or resources, certainly they would want to make sure that their evil AGI is keeping up with the Joneses.

For a brief period, the good AGI is slightly ahead of the evil AGI in terms of AI features. The evildoers catch up. The evil AGI and good AGI are now evenly matched. As you might guess, the good AGI might subsequently and reactively be further advanced by the good intended people. The good AGI is once again ahead of the evil AGI.

Rinse and repeat.

An endless battle of AI advances could take place. Unless one of those advances is a knockout blow to the other AGI, the tug-of-war or cat-and-mouse gambit will just keep plugging along.

• 14) Suppose the AGI police come up with or can utilize a good AGI to conquer the evil AGI, but then opt to deploy the good AGI as a means of suppression

Let’s not forget the stipulation about potentially putting together an AGI police force. This would be humans that police the world using the presumed good AGI.

Some would wonder whether the AGI police might get ahead of themselves. Perhaps they use the good AGI as a means of suppression or enslavement of humans all told. Why would they do this? Maybe in the name of saving the world from the evil AGI. Perhaps great power corrupts greatly. Could be lots of proffered reasons.

You might have an earnest belief that the good AGI would assuredly not let the AGI police get away with subverting the purpose of the good AGI. The good AGI is good. Unlike a “dumb” weapon that has no semblance of how it is being used, we would assume that the good AGI understands what is taking place. Perhaps the good AGI would not tolerate being diverted to unsavory purposes.

Makes you feel warm and fuzzy that the good AGI might want to save us from ourselves.

Conclusion

I can tell you right now that some will howl and denigrate all this talk about evil AGI and good AGI. It is the stuff of sci-fi stories and not fitting to be given serious consideration, they will exhort.

Just all preposterous stuff.

Some disagree.

Elon Musk has repeatedly stated that AI could lead to civilization’s destruction.

Bill Gates has stated in his online blog that:

• “These ‘strong’ AIs, as they’re known, will probably be able to establish their own goals. What will those goals be? What happens if they conflict with humanity’s interests? Should we try to prevent strong AI from ever being developed? These questions will get more pressing with time” (“The Age of AI Has Begun” by Bill Gates, the online blog of March 21, 2023).

A final thought on this topic for now.

The famous novelist Robert Louis Stevenson said this about humankind: “All human beings are commingled out of good and evil.”

Will AGI be of a like intertwining?

Can we prevent any such intertwining and devise only good AGI?

Could we overcome an entirely evil AGI?

These are substantial questions with unresolved answers. We might be wisest to resolve them or at least closely address them, sooner rather than later, and preferably before we arrive at AGI of any flavor or derivation. Better to be safe than sorry, as they say.