Thoughts on GenAI

GenAI seems to have become the tech world’s latest “shiny object” that will solve all the world’s problems.? I see it differently.? I? commonly get asked, “What do you think about GenAI?”.? The question seems simple but has many complex undertones.

Tech

Technology is only one part of the solution, and it is the easy part: GenAI uses large language models(LLM), which are, in turn, part of a field of study known as Machine Learning. These LLMs involve the advanced use of statistical techniques and massive mathematical computations. These techniques are awe-inspiring and robust and are still a topic of intense research.? They are very real but don’t match the hyperbolic proclamations of the snake-oil salesmen.

Like all tools, they have to be used appropriately.? In the early days of the field, the Stochastic Parrot (https://en.wikipedia.org/wiki/Stochastic_parrot) metaphor was used to describe the weaknesses of LLMs.? LLMs do not “understand” or “reason” in any absolute sense.? The level of “truth” in LLMs is statistical.? While this criticism is still valid, much work has been done to get beyond statistics.? There is still much work to be done.

Statistical models function based on training data, which is digested into extensive statistical data sets that reflect the training data. If the training data contains unfiltered social media postings, misogyny, and racism (amongst many other misbehaviors), these components will be baked into the model.? If the training data includes conversations involving lawyers and law students, the GenAI tool will probably be very good at producing answers to LSAT questions.? However, answering LSAT questions isn’t the same as being a lawyer.

Artificial Intelligence has a long history (https://en.wikipedia.org/wiki/Timeline_of_artificial_intelligence), even though the term wasn’t coined until 1956. We don’t even have a straightforward way to test what would genuinely be a competitor to human intelligence. In 1950, Alan Turing proposed what has come to be called the Turing Test(https://en.wikipedia.org/wiki/Turing_test): if you can have a conversation with the candidate intelligence and not be able to tell whether it’s human, then it’s real AI.? The flaw in the Turing test is that humans are gullible.? In 1965, Joseph Wizenbaum wrote a program called Eliza (https://en.wikipedia.org/wiki/ELIZA), a simple simulation of a psychiatrist.? Many people interacted with Eliza, and it took a while for many of them to realize that there was no actual human on the other side.? Even after finding out what Eliza was, people often continued the conversation because they found it helpful.? Text conversations with a GenAI system feel like conversations with a turbocharged Eliza, which they essentially are.? However, as your questions get more complex,? the probability of an error in the answer rises, and sometimes the erroneous answers can be hilarious.? Sometimes, the errors can only be discerned by an expert.? GenAI is remarkably useless in handling challenging problems, but they’re fine for things like a first-level support line (something like a weak Google search that only gives one answer).

Hype

The term AI has been attached to many technologies over the years. Non-specialists always overestimate its capabilities because what pops into their minds is decades of fiction in movies and books: a general intelligence superior to humans. The media and analysts get all breathless about AI, and many people, including C-suite folk, get caught up in the whirlwind. Sometimes, the C-suite folk know better but keep getting trapped by the media and analysts for “a GenAI strategy.”

Then there are the Venture Capitalists.? There’s this mythology that they’re driven by innovation.? The truth is that their primary goal is (unsurprisingly) money.? This causes them to mindlessly follow the trendy fashions of the day to attract investment.? Then, they fiercely pursue their exit (https://www.google.com/search?q=vc+exit) path to cashing out, which only accidentally leads to leaving a healthy company behind.? They’ve been pouring money into GenAI firms like crazy lately.? VCs are such copycats.? They slacken their due diligence to follow the fashion trend.

AI has repeatedly blown up in hype cycles, but it happens with other tech, too. Cryptocurrency is the most dramatic recent example, and it had the bonus craziness of being both a Ponzi scheme and a channel to launder money. GenAI is far more respectable, but folks unencumbered by morals can find a way to exploit anything.

Fortunately, the hype is starting to get toned down.? The breathless marketing claiming that GenAI will solve all the world’s problems has started to back off.? I even see the occasional article mentioning the possibility of a bubble.? Of course, it’s a bubble.? But it’s still pretty intense (https://www.washingtonpost.com/technology/2024/08/13/ai-ads-olympics-commercials-overload-google/).

Apocalypse

This technology hype cycle has featured all kinds of speculation that GenAI will become all-powerful and be the downfall of human civilization.? That’s just silly: it’s like saying that having the ability to multiply huge matrices will end humanity.? If AI is involved in ending humankind, it will be because some idiot human decides that it would be a good idea to build armed drone aircraft, give them autonomous kill authority, and an automated repair & refuel facility.? We’ve seen this movie.? Strangely enough, I once had someone pitch me to join a team to build this.? They put nice assuring words around it, but it was SkyNet.

A lesser but more likely apocalypse is what commonly happens to AI: the bubble bursts and the underlying tech is painted as the cause of the hype’s failure and is dumped in the trash. ? GenAI has very good uses, which we shouldn’t abandon.

The more immediate apocalypse is what’s happening to everything that is not GenAI: funding is vanishing. VCs, big tech companies, and anyone who’s a customer of them are steering investment toward GenAI. Engineers have learned quickly that if they want to get a project funded, they need to sprinkle some GenAI dust on it. This is nuts, but humans know how to survive, even if it involves gaming the system.

Perhaps the saddest apocalypse is when the marketing folks start pitching, and senior managers start hearing, “Think of all the people you can lay off!"

When the GenAI/LLM hype bubble bursts, I hope the tech doesn’t die, too. It’s pretty cool when used appropriately.