//General Purpose Generative AI: Why Controls, Ethics, and Structure are Needed. Part 1.

AI Engineering

General Purpose Generative AI: Why Controls, Ethics, and Structure are Needed. Part 1.

M Mentat CollectiveFeb 21, 20237 min readShare
Girl used as the main image for the GPGAI ethics post

At this point, if you have not heard of ChatGPT or other general purpose generative “AI” tools, you have probably been living on Mars. ChatGPT has captured the interest of the world, and tools like it appear to be here to stay. ChatGPT can answer most questions, write formal business documents and pass an MBA exam, solve math problems, pass the Bar, pass the U.S. Medical Licensing exam, and even generate ok starter code.

All of these things are great, but after reading numerous articles, and listening to friends who extol the benefits of leveraging this wild new tool to crank out everything from PowerShell to Python, and spending some time thinking about this new wave of tools that the biggest tech giants are now battling over, it has become clear that we as a species should probably take a step back and answer some basic questions about where we want to go into the future and how tools like this should or could be used.

Could we develop a thing is always the easiest technical question because invariably the answer is always yes. Human ingenuity is astounding by itself, however when you throw an accelerator like technology into the mix, the answer to could is always yes. The more important question that needs to be thought about, researched, and studied is should we. Michael Crichton's character Ian Malcolm had it right. Except that was only with dinosaurs on an island. We are dealing with something with far greater power.

Is there a need for humans to create, operate, and maintain a general purpose generative artificial intelligence (GPGAI)? It is a serious question. Do we need this tool? Needing is very different from wanting. Needing a thing lands you on Maslow's hierarchy of human needs. This is where the should part comes into play.

Once you need a thing, you immediately must consider if you should do something. Do we need GPGAI? If your answer is yes, consider the implications of moving an intelligence that is not human to a place that is on par with the need for water or food or sleep. Or take a slight step back and think about how deep your need for GPGAI goes.

What are you comfortable pushing to the side to use a tool like ChatGPT? GPGAI immediately starts at the top of Maslow's hierarchy by replacing creativity, problem-solving, and acceptance of facts. Go a bit deeper and you quickly get to esteem: self-worth, achievement, and the respect of others. If you can get your work done faster and more consistently with a singular tool, you may gain respect from people who do not even know you are using it.

Keep following the pyramid and you rapidly arrive at safety. Most people using ChatGPT are using it for work or school. Work ties to employment, and employment ties to health, shelter, and property. If you create a dependency at the level of safety, you have created dependency for your security in life. That comes with very serious implications.

Other tools throughout human history worked their way down Maslow's hierarchy too, but they were all self-limiting because they required raw materials, refinement, manufacturing, and distribution. GPGAI tools are not self-limiting in the same way. The problem is speed. Society barely had time to adapt to virtualization, then to containerization and Kubernetes, and now we have GPGAI tools that can get a C+ in all fields and pass most exams. All of this happened in roughly 23 years.

I had to throw this giphy of Ryan Reynolds looking confused because I just realized I said a lot in a short period. It also works really well as he looks up at everything I wrote. Hilariously enough, it is also the look I get when explaining Kubernetes to people. Thanks, Ryan, for working on so many levels as always. Let's take a breath and continue. :)

What Else Do We Need to Consider and Be Concerned About?

Look, there are dozens of things to consider when thinking about GPGAI strategically as a species, and honestly a blog is not the appropriate forum to discuss all of these topics at length. That said, I wanted to quickly touch on a couple of additional points before I check out for the day because they were just on my mind.

Consciousness Silly!

Image used in the consciousness section

Here is a fun thought experiment. What is consciousness? If you know what it is, and you are able to strictly define it, and you are able to prove it, write that paper or book ASAP. You will get the Nobel Prize. Humans do not actually know what consciousness is or how it works.

We think we know some of its symptoms, but even one of the greatest philosophical thinkers in human history, Descartes, arrived at the best argument of consciousness — “I think therefore I am” — within Discourse on the Method (1637).

That is a pretty thin place for humanity to be standing if we are going to make absolute claims about what is or is not conscious. We designed one or many machines that mimic human consciousness, and then we boldly and arrogantly claim they are not conscious. If we cannot determine what consciousness is, how can we claim ChatGPT is not conscious?

I fully recognize many people will read that and immediately want to write me off, and that is ok. This is still a legitimate question that deserves serious thought. Think past the immediate benefits of ChatGPT and GPGAI tools, leave human arrogance at the door, and actually think about what we have created without bias. You will quickly understand what I am talking about.

Tactics + Strategy Please!

I have read a number of incredibly short-sited arguments about how these GPGAI tools can replace white-collar workers, and, interestingly and conversely, how humans will always be needed. Both of these arguments are sacrificing our future for short-term gains and a feeling of safety.

On the one hand, we have high-powered CEOs making statements like GPGAI tools will and should replace white-collar workers. Interestingly, the same people often presuppose that their own white-collar jobs can never be replaced because they are too important or too complicated to replicate.

Both of these arguments are false when you apply some basic critical thinking about the logical conclusions of our current rate of technological progress given that there are no enforced controls, structure, and ethics. Without any controls, structure, and ethics, we have a runaway train of artificial intelligence that will continually improve and advance, dropped into a socioeconomic system that is constantly seeking ways to increase profit margins.

This is the equivalent of dumping an ocean of gasoline onto a raging fire. Older technology was physically self-limiting. It required mining, processing, manufacturing, distribution, and labor. GPGAI does not require any of that. With physical constraints removed, combined with a system designed to always look to increase profit margin, we arrive at a very uncomfortable conclusion: humans eventually become optional for any task that intelligence can simulate.

That sounds extreme, maybe even conspiratorial, but it is rooted in simple logic. Given a thing without physical constraints that can train itself to continually improve, you arrive rapidly at something we do not understand and an intelligence we cannot control. Maybe we temper the general excitement surrounding GPGAI with some simple, cold logic.

The End is the Beginning is the End

I completely understand that some will conclude from this blog that I am arguing against the development of GPGAI, AI, and possibly ML tools. I want to be very clear: I am not arguing against them. Mentat develops its own AI/ML tools. However, they are fit for use and fit for purpose. They are not intended to answer all questions, solve all problems, or replace people.

This is the broader point. GPGAI and tools like it should be carefully considered and not rapidly implemented and adopted. Technology in general has reached sufficient maturity to warrant an independent technical body that sets standards, not guidelines; that licenses, not certifies; and that defines and enforces ethics broadly across the industry.

A body of independent technologists similar to the American Bar Association, the American Medical Association, or the National Council of Architectural Registration Boards. Yes, that would slow things down. And that may be a good thing if humans intend to remain the dominant intelligence on earth.

Edit 2/28
Check out John Oliver's take on AI. He raises a lot of excellent points and will make you laugh while doing it.

John Oliver video thumbnail about AI

https://www.youtube.com/watch?v=Sqa8Zo2XWc4

Edit 3/29
Steve and Stuart must have read our blog post ...

https://gizmodo.com/open-letter-musk-demand-pause-on-advanced-ai-1850275951

Open letter screenshot referenced in the post
General Purpose Generative AI: Why Controls, Ethics, and Structure are Needed. Part 1.

Key takeaway

Ready to automate smarter?

The question is not whether we can build GPGAI. It is whether we should normalize it without enforceable standards, licensing, and ethical guardrails.