ChatGPT As A Platform Gets Bigger And Bolder As OpenAI Rolls Out Plugins For All Kinds Of Add-On Apps, Stewing Up AI Ethics And AI Law

If you are keeping tabs on the AI realm, you might have felt a shudder or powerful disturbance in the Force this week.

I am referring to the fact that OpenAI announced and released on Thursday, March 23, 2023, the capability to have plugins associated with their widely and wildly successful generative AI app known as ChatGPT. This has undoubtedly caused a ripple in the energy force across all of the AI realm and beyond.

In today’s column, I will explain the significance of the plugins and how they will stoke even more attention to ChatGPT. The giant seems to keep on getting bigger. The air in the room is being consumed by OpenAI and ChatGPT. Other generative AI makers and their apps keep trying to go toe-to-toe, see for example my coverage at the link here, meanwhile, the OpenAI speeding train amazingly continues to go faster and faster. Unimpeded, undeterred. All speed ahead is the existing mantra.

For my ongoing coverage of generative AI and the latest twists and turns, see the link here.

I’m sure that you already know that ChatGPT is a headline-grabbing AI app that can produce fluent essays and carry on interactive dialogues, almost as though being undertaken by human hands. A person enters a written prompt, ChatGPT responds with a few sentences or an entire essay, and the resulting encounter seems eerily as though another person is chatting with you rather than an AI application.

Please know though that this AI and indeed no other AI is currently sentient. Generative AI is based on a complex computational algorithm that has been data trained on text from the Internet and admittedly can do some quite impressive pattern-matching to be able to perform a mathematical mimicry of human wording and natural language. To know more about how ChatGPT works, see my explanation at the link here. If you are interested in the successor to ChatGPT, coined GPT-4, see the discussion at the link here.

Consider these four primary modes of being able to access or utilize ChatGPT:

  • 1) Directly. Direct use of ChatGPT by logging in and using the AI app on the web
  • 2) Indirectly. Indirect use of kind-of ChatGPT (actually, GPT-4) as embedded in Microsoft Bing search engine
  • 3) App-to-ChatGPT. Use of some other application that connects to ChatGPT via the API (application programming interface)
  • 4) ChatGPT-to-App. Now the latest or newest added use entails accessing other applications from within ChatGPT via plugins

When ChatGPT was first starting out back in November, I had predicted that the third item listed above would gradually and inevitably bolster the use of ChatGPT to heightened levels of everyday usage, see my written analysis at the link here. The availability of an API was a smart move and meant that more than just direct usage would arise. In short, other software makers could opt to leverage ChatGPT by connecting their software to ChatGPT. This is undertaken by a programming convenience known as an API. Essentially, allowing one program to access another program.

For example, suppose a maker of a sales package wants to allow their users to generate snazzy emails for sending to sales prospects. Rather than trying to program or code such a capability, all they need to do is include a portion that makes use of the API connecting to ChatGPT. Voila, your software package suddenly can do all manner of Natural Language Processing (NLP).

All told, this meant that the uses of ChatGPT could expand immensely.

Firms that provide apps were thirstily lured to the ChatGPT API like a siren call. You might ostensibly declare that it was nearly a no-brainer to proceed to get connected. The special bonus was that the software maker could then tout that they too are a card-carrying member of the heralded ChatGPT wonderment club. One cheerful consideration proffered the obvious point that adopting ChatGPT usage could enhance the features of your software. The other and especially vibrant plus is that your software gets that oh-so-awesome ChatGPT afterglow.

This is a bandwagon that is certainly good for the other allied software makers and equally good for OpenAI and their persistent and outsized efforts to garner larger and larger usage of ChatGPT (a seemingly never-ending quest).

Let’s step up the game and add plugins into the mix. The bandwagon has now gotten even more attractive. There will be a lot more interest in further piling into the ChatGPT bandwagon. All due to the addition of plugins.

Try this on for size: ChatGPT as a platform.

Some are saying that the plugins will inexorably and dramatically shift ChatGPT into the enviable position of becoming ChatGPT as a platform. Whereas the up-until-now API connections meant that other apps would invoke ChatGPT, the hot and heavy facets of plugins are that this means you can now use other apps while making use of ChatGPT.

Look at it this way, as though driving on a street:

  • a) Driving down the street via App-to-ChatGPT. You could perchance use someone else’s app that happened to have set up a connection to ChatGPT via the API
  • b) Driving up the street via ChatGPT-to-App. Now you can be using ChatGPT and have it invoke some outside app that provides complementary capabilities or useful augmentations for ChatGPT

Both directions are possible now.

They are not mutually exclusive of each other. Going back to the sales package example, a user in the sales package might make use of ChatGPT while inside the sales app. Maybe they request a bunch of emails is composed for a prospects list. That would be the already conventional App-to-ChatGPT routing.

The plugins add the following switcheroo direction. You are in ChatGPT. You want to use a sales package. Normally, you might have to switch back and forth between using ChatGPT versus using the sales package. Instead, assuming that the maker of the sales package sets up a plugin for ChatGPT and gets it approved by OpenAI, you can simply ask ChatGPT to use the sales package. This could also be automatically done for you, depending on how you’ve got ChatGPT set up. In any case, this is the newly made available ChatGPT-to-app routing via plugins.

Notice that the sales package can have both options available. Users inside the package are able to leverage ChatGPT. Meanwhile, people using ChatGPT are able to use the sales package. Convenience for everyone involved.

Those using ChatGPT might not have ever thought to use the sales package, maybe not even aware of its existence, but now they can readily do so via the push of a button or by giving a quickly written instruction to ChatGPT to do so. The sales package is almost abundantly going to land a new base of users. Furthermore, similar to the earlier point about garnering the ChatGPT afterglow, there is going to at least initially be a bit of prestige about being a ChatGPT plugin.

Imagine the marketing team of the sales package maker going bananas and celebrating this. The sales package maker can gleefully yell to the rooftops that they are honored to be accessible from ChatGPT. They must be a reputable company. The trust and faith in the wonders of ChatGPT are going to bring fame and stardom to the sales package maker.

Bonanza to be had.

App makers of all kinds should be salivating over this grand possibility. All they need to do is develop their plugin, submit it for consideration, and if approved by OpenAI they are now added to the esteemed world of ChatGPT. If they already had earlier done an API connecting their app to ChatGPT, this is pretty much like falling off a log to also craft a plugin.

Wake up and smell the roses. Software firms are going to rapidly attempt to become part of the ChatGPT platform, as it were. Pundits have been clamoring that this is akin to the phenomena that happened when the iPhone made available the SDK (software development kit) for devising apps for the iPhone. The iPhone was the platform. The plethora of apps that were created made the iPhone increasingly useful.

A type of cycle takes place in these circumstances. Sometimes referred to as a network effect, see my analysis at the link here, people tend to join something that others are joining. Facebook was this way. Snapchat was this way. At first, maybe there is little or no traction. But, then, often out of the blue, people start to join. Their friends and colleagues join. Everyone wants to join.

The big get bigger. The small get starved or fail to get any oxygen in the room. That’s the gist of the network effect. It becomes a form of stickiness to the exponential growth factor. People will use what everyone else is using. This in turn makes it more alluring and adds value. The snowball is at times unstoppable and gathers erstwhile momentum.

Can ChatGPT go into the slipstream or tailwinds of the network effect and become the next really big thing that everyone aims to use?

Yep, sure can.

I will unpack this exciting topic next.

As a trigger warning, realize that there are some potential downsides to all of this. The mighty can potentially reach new heights. In the same breath, keep in mind that the mighty can fall from high perches. I want to highlight the good, the bad, and the ugly that might be further down the road.

On the upbeat side, here’s what OpenAI’s CEO Sam Altman stated in his tweet announcing the rollout:

  • “We are starting our rollout of ChatGPT plugins. You can install plugins to help with a wide variety of tasks. We are excited to see what developers create!” (posted on March 23, 2023).

The next day, Elon Musk tweeted on March 24, 2023, in response to a tweet regarding concerns that the release of the ChatGPT plugin capability might be hasty regarding crucial AI safety considerations, stated this in his usual straightforward and no-nonsense way:

  • “Extremely concerning.”

You see, we are facing the best of times, or we might be facing the worst of times. It all depends.

Into all of this comes a slew 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.

I’ll be interweaving AI Ethics and AI Law related considerations into this discussion.

The ChatGPT Plugins Come Calling

I’ll give you a quick guided tour of what the ChatGPT plugins are all about. During this tour, I will point out significant facets, including what is good and what perhaps might raise some wary eyebrows.

First, let’s see what OpenAI says about the ChatGPT plugins (source: OpenAI official webpage on Chat Plugins).

  • “OpenAI plugins connect ChatGPT to third-party applications. These plugins enable ChatGPT to interact with APIs defined by developers, enhancing ChatGPT’s capabilities and allowing it to perform a wide range of actions.”
  • “Plugins can allow ChatGPT to do things like:”
  • “Retrieve real-time information; e.g., sports scores, stock prices, the latest news, etc.”
  • “Retrieve knowledge-base information; e.g., company docs, personal notes, etc.”
  • “Perform actions on behalf of the user; e.g., booking a flight, ordering food, etc.”

Let’s analyze the indications.

In case you didn’t already know, a notable limitation of ChatGPT has been that it could not readily access new data in real time from the Internet. The data training for ChatGPT was concluded in 2021. Thus, any questions or topics that you bring up with ChatGPT will not encompass the year 2022 or the year 2023. That’s a bummer, for sure.

But the world has now changed, well, somewhat.

They aren’t overtly retraining ChatGPT and instead are just allowing access to more up-to-date data. This is both handy and yet also somewhat disappointing. On the one hand, you can have ChatGPT look up stuff on the Internet and react to it when composing essays or responding to your interactive conversation. Hurrah. Many users of ChatGPT will be enormously excited about this new capacity.

That doesn’t though mean that ChatGPT will necessarily later on recall or have templated whatever the discussion entailed. Think of it this way. You use ChatGPT and in so doing have it look at data from the Internet that indicates there was a sizable snowstorm in Northern California in March 2023. ChatGPT converses with you about the snowstorm. All seems good. Later on, someone else that uses ChatGPT asks about the snowstorm, but ChatGPT might indicate that it doesn’t have data about the claimed snowstorm since it is beyond the data lock of 2021.

I’m sure you can see why this is not quite like eating cake and having the icing too. The icing isn’t there.

Another important point mentioned in the above indication was the idea that you can retrieve data such as company documents, personal notes, and the like.

With conventional ChatGPT, you had to feed any documents or notes into ChatGPT by pretty much grabbing the text from those materials and pasting them into the ChatGPT prompt window. This was a pain in the neck. Some AI developers right away created programs that would allow you to submit a file such as a PDF and their program would feed that text into ChatGPT for you via the API. Handy. Turns out though that with the new addition of plugins, those separate programs have become essentially disrupted or to some extent obsolete (they can still be used, but probably the plugins will dominate such usage).

Time moves on, a classic and harsh rule-of-thumb that confronts all developers of software.

There is something else about the notion of feeding company documents and personal notes that the plugins are potentially endangering, namely privacy and data confidentiality. I’ve covered the qualms that people were mindlessly feeding company proprietary info into ChatGPT at the workplace, doing so without realizing that OpenAI licensing clearly warns that there is no semblance of data confidentiality. Same with entering your private info. There isn’t any ironclad protection for the privacy of your entered data. See my coverage at the link here.

The worrisome point here is that with the ease of plugins, the chances of making those privacy intruding mistakes and the chances of violating data confidentiality are bound to go up. Likely multifold. That being said, if the makers of the plugins have their act together, they will emphasize in various alerts and cautions that the user should not let themselves compromise the data that they have.

The official webpage of OpenAI that covers the ChatGPT plugins says this:

  • “The open-source retrieval plugin enables ChatGPT to access personal or organizational information sources (with permission). It allows users to obtain the most relevant document snippets from their data sources, such as files, notes, emails or public documentation, by asking questions or expressing needs in natural language.”

A key element consists of the phrase “with permission” and I can only hope that this is taken seriously and with great caution. From an optimistic sunny side perspective, maybe this will reduce the ad hoc efforts that are already falling into this trap. Ergo, this helps to solve an existing problem. Taking the pessimistic perspective, one might say that this exacerbates the problem because it will become easier to be sparked into making privacy and confidentiality mistakes.

You judge.

Shifting gears, you might be wondering what the status of the plugin’s capabilities is.

Good question.

Here’s what the OpenAI official webpage on ChatGPT plugins indicates:

  • “Users have been asking for plugins since we launched ChatGPT (and many developers are experimenting with similar ideas) because they unlock a vast range of possible use cases. We’re starting with a small set of users and are planning to gradually roll out larger-scale access as we learn more (for plugin developers, ChatGPT users, and after an alpha period, API users who would like to integrate plugins into their products). We’re excited to build a community shaping the future of the human-AI interaction paradigm.”

The crux of that indication is that they are starting the plugins in a limited manner. Some would applaud this approach. Better to dip a toe into these uncharted waters rather than diving in completely. The counterargument is that they ought to be doing this in a private setting and not with the general public per se. In other words, hire or pay people to try out this stuff. Keep it behind the scenes. When everything seems to be checked out, go ahead and launch.

Should generative AI such as ChatGPT be utilizing the techie-held belief of “move fast and break things” when it comes to rolling out AI capabilities?

You could insist that this is readily accepted and the best path to innovating via AI. The contrary view is that this is making us all into guinea pigs in an experiment carried on by AI makers. Recall that Elon Musk seemingly expressed a view that the ChatGPT plugins rollout is “extremely concerning” pertaining to AI safety via his recent tweet.

For those of you eager nonetheless to get your hands on the vaunted plugins, here’s what the OpenAI webpage on the ChatGPT plugins indicates about which plugins are available at this time:

  • “Plugin developers who have been invited off our waitlist can use our documentation to build a plugin for ChatGPT, which then lists the enabled plugins in the prompt shown to the language model as well as documentation to instruct the model how to use each. The first plugins have been created by Expedia, FiscalNote, Instacart, KAYAK, Klarna, Milo, OpenTable, Shopify, Slack, Speak, Wolfram, and Zapier.”

I suppose your eyes immediately went to the list of firms named at the end of that quotation.

The plugins, when available for use, so far include that you can use ChatGPT to help plan a trip and then use say Expedia to do the bookings and/or use KAYAK to help with flights and rentals (some of those are overlapping capabilities). Ordering from your local grocery store can be done while in ChatGPT and then leverage the plugin for Instacart. When you want to use ChatGPT to help decide what to have for lunch or dinner or whatever, you could then use the plugin to OpenTable to book a restaurant for the scrumptious meal. And so on.

How will you know which plugin to use for a particular circumstance at hand?

When the number of plugins increases, it would seem that there is not necessarily going to be a dedicated turf or territorial arrangement regarding the granting of plugins. You might have assumed that there would be perhaps one of each type of occasion or category, such as one and only one plugin for a chosen travel agency type of service, and one and only one plugin for a chosen online shopping service, etc.

Instead, it appears that the viewpoint seems to be that the more the merrier. Let the user decide from a wide variety of choices, even if they overlap in the services provided. Survival of the fittest. No need to decide in advance. I’m sure that some will hail this approach. Others will carp that it is going to potentially be confusing to the users of ChatGPT.

One also wonders what the marketing campaigns are going to be like. Imagine that your firm has established a ChatGPT plugin for providing your respected dog-walking services. Suppose there are fifty other such plugins for the same exact service. You want your plugin to be chosen over the others. This might be a boon for marketing pros that can figure out how to convince people to use a particular ChatGPT plugin over another one, of course as being paid to do so by the firm that makes or profits from the particular plugin.

Get those mind wheels spinning.

I will cover a few more ChatGPT plugin facets and then wrap things up with a short conclusion.

Here’s what the OpenAI webpage about the ChatGPT plugins says about two very special plugins, hosted by OpenAI rather than a third party, and consisting of (1) a web browser plugin, and (2) a code interpreter plugin:

  • “We’re also hosting two plugins ourselves, a web browser and code interpreter. We’ve also open-sourced the code for a knowledge base retrieval plugin, to be self-hosted by any developer with information with which they’d like to augment ChatGPT.”

I already mentioned the web browser plugin when I was discussing the ability to access now the Internet while inside ChatGPT, which as I said you could not readily access before (other than, kind of, via Microsoft Bing).

You might be puzzled as to why ChatGPT has not already made readily available access to the Internet. I had earlier emphasized that one form of access entails ChatGPT getting additional data training, while a different form of access entails merely looking stuff up for dealing with a particular conversation underway.

Focus for a moment on the latter, consisting of merely accessing the Internet for a conversational purpose and not for a data training purpose.

We all know that the Internet is a vast source of incredible information that can be tremendously helpful and serve as a boost to sharing knowledge around the world. Score one point for the Internet. The Internet can also contain the worst of the worst. There is misinformation. There is disinformation. There is patently offensive material. The Internet can be a sewer. Nobody can dispute that. Deduct one point for the Internet. Up one, down one.

Here’s why that’s important regarding ChatGPT.

All manner of refinement was done by OpenAI to ChatGPT before they released ChatGPT to the public at large. For example, the use of RLHF (reinforcement learning via human feedback) is a significant technique for trying to get generative AI to avoid producing offensive essays or repeating false facts that were garnered during the data training stage. Human reviewers are shown various ChatGPT essays and they rate the essays to provide feedback to the AI algorithms. The algorithms try to pattern-match what is considered acceptable to then output versus what is unacceptable. See the link here for my details on this approach.

By then freezing the data training at the date lock of 2021, there was a somewhat reasonable chance of trying to fend off the AI pattern-matching from generating horrendous outputs. To some degree, this was successful in presumably reducing the frequency of such unseemly essays, though they are still produced and you can spur ChatGPT to do so, see my coverage at the link here of the hate speech and other untoward essays that people have gotten out of ChatGPT.

So the upshot is that without real-time Internet access, there was a presumed reasonable odds that ChatGPT would not spout off with crude language. The fact that OpenAI is now providing a web browsing plugin seems to fly in the face of that earlier precaution.

Another twist is that ChatGPT could potentially do active things on the Internet if the AI app has unfettered access to the Internet. For example, suppose you want to sign-up for a subscription to your favorite magazine. You could potentially ask ChatGPT to do so for you, assuming that a web access facility had been included. ChatGPT could access the magazine website, fill in a subscription form on your behalf, agree to whatever pricing terms or licensing might be there, and then inform you that you are now signed up.

Some believe that allowing AI apps to actively do things on the Internet is a dangerous slippery slope.

In the case of the magazine subscription, suppose ChatGPT inadvertently signed you up for the wrong magazine or perhaps signed you up for a hundred magazines. Yikes. Do we want AI acting on our behalf and taking actions that might be erroneous or have adverse consequences?

The next thing you know, we’ll have AI doing all manner of crazy things on the Internet. Out of this, we might eventually find ourselves facing a grievous existential risk. I’ve discussed the ongoing debates and controversies about whether we are setting ourselves up for having AI harm or wipe out humanity, even if AI doesn’t reach sentience (the emphasis being that AI sentience is not the only point at which we are potentially endangered), see the link here.

Do you think that generative AI should or should not have real-time access to the Internet?

While you are pondering that hefty question, take a look at what the OpenAI webpage about the ChatGPT plugins mentions on the topic of the web browser plugin:

  • “We’ve created a web browsing plugin which gives a language model access to a web browser, with its design prioritizing both safety and operating as a good citizen of the web. The plugin’s text-based web browser is limited to making GET requests, which reduces (but does not eliminate) certain classes of safety risks. This scopes the browsing plugin to be useful for retrieving information, but excludes “transactional” operations such as form submission which have more surface area for security and safety issues.”

Notice that the web browser plugin is said to have a read-only restriction. You can bet that some hackers will try to find ways to break out of that restriction. Also, the chances are that other plugins are bound to go beyond a read-only and have some form of Internet access that isn’t just limited to GET requests.

There’s that slippery slope consideration too.

Finally, I cover one last aspect for now in this discussion.

You might be wondering how you will make use of the plugins while inside ChatGPT as a user.

According to the OpenAI webpage on the ChatGPT plugins, here’s the deal:

  • “When starting a conversation on chat.openai.com, users can choose which third-party plugins they’d like to be enabled. Documentation about the enabled plugins is shown to the language model as part of the conversation context, enabling the model to invoke appropriate plugin APIs as needed to fulfill user intent. For now, plugins are designed for calling backend APIs, but we are exploring plugins that can call client-side APIs as well.”

This techno chatter boils down to the idea that you will be able to selectively decide which plugins to enable and which ones to keep dormant. This will apparently be selectable on a user-by-user basis. Skeptics wonder if at some point there will be plugins that are automatically enabled by default, such as if a plugin maker pays a special fee to ensure that their plugin is always active.

One never knows what the future might hold.

Conclusion

There is a lot more that I could say about the ChatGPT plugins, but I’m nearing the space limitations herein and will save those remarks and insights for upcoming column postings. Be on the look for more explorations of the plugins and the mania likely to be arising about them.

Let’s end this saga with a bang.

You might vaguely know that generative AI such as ChatGPT has many flaws. Besides the possibility of producing offensively worded essays and interactions, there are many additional and extremely disconcerting issues about today’s generative AI.

Four concerns about generative AI that I have extensively covered include:

  • 1) Errors. Generates wording and essays that have errors of fact or miscalculations, etc.
  • 2) Falsehoods. Generates false assertions and other insidious falsehoods.
  • 3) Biases. Generates wording and essays that contain biases of nearly any and all kinds.
  • 4) AI Hallucinations. Generates what appears to be factual but is made-up and not at all factually based (I don’t like the term “AI hallucinations” due to the anthropomorphizing of AI, but it seems to be a catchphrase that has regrettably gained acceptance).

Lest you shrug off those pitfalls, realize that people using generative AI are bound to fall into the trap of accepting the outputted essays as truthful and factual. Doing so is easy-peasy. You see lots of essays and interactions that seem on par with human levels of fluency and confidence. You get lulled into assuming that everything uttered is of the utmost correctness.

Even the most ardent supporters of generative AI would acknowledge that we have severe problems associated with the generation of errors, falsehoods, biases, and AI hallucinations. No reasonable AI researcher or AI developer could disagree with that contention.

Why do I bring this up?

Because the added use of plugins and the potential for ChatGPT to be construed as and become a platform unto itself means that we are also establishing a foundation that we already know consists of or can generate errors, falsehoods, biases, AI hallucinations, etc. This is a sailing ship that already has lots of holes and a sail that is hazardously marred.

Critics would say that we ought to not be launching this sailing ship at this time. You are going to put us all into this ship as it attempts to sail around the world. As I earlier pointed out, the chances are that lots and lots of software makers are going to jump on the ChatGPT plugin bandwagon. We will have a grand interdependency between zillions of everyday apps and ChatGPT.

The question you’ve got to ask yourself is whether this is the right time and place to have a ship that is aiming to be our flagship, and for which due to the existing leaks and maladies, it might at some point capsize or take us down with it.

There is a tradeoff to all of this, of course. You could try to claim that the benefits of generative AI outweigh the downsides. You have to crawl before you walk. You have to walk before you run. We have to start someplace, goes the typical refrain.

A final word or two of a philosophical bent might be worth contemplating about this quandary.

Christopher Marlowe, the famed English playwright, said this: “What nourishes me, destroys me.”

Let’s work together and plug in our collective energies to ensure that generative AI is altogether nourishing, meanwhile averting the foreboding endangerment about being able to destroy us.

That seems like a pluggable and worthwhile goal to me.

Source: https://www.forbes.com/sites/lanceeliot/2023/03/25/chatgpt-as-a-platform-gets-bigger-and-bolder-as-openai-rolls-out-plugins-for-all-kinds-of-add-on-apps-stewing-up-ai-ethics-and-ai-law/