AGI: Pipe dream or the future?

With every leap in data querying technology, there is renewed excitement about Artificial General Intelligence (AGI). So it was no surprise that in 2023, when OpenAI released ChatGPT for general use, it got us all thinking about Tony Stark’s Jarvis and how we would all have AI assistants to be our constant companions. Two years later, we are still trying to figure out how AI fits into our world and, more importantly, into our products. I even saw an ad for a body cream that claimed to use AI. We see elements of AI injected into almost every software product - some effectively, while most are just window dressing at best and irritating at worst. For instance, as I type out this article, Gemini is constantly trying to guess my next word. At times, it’s helpful. But at other times, it’s just plain irritating since the suggested word is close enough to send my thoughts spinning in a tangent. 

Of course, it is the holy grail. We dream of developing an intelligence that thinks like us, is as creative as us - just faster. But is it possible to create such a tool? Personally, I believe AGI will never happen. Let me lay out my case. 

The biggest limitation to developing AGI is the limits of our imagination. Human minds struggle with randomness and time. Let’s take randomness first. Imagine us looking at the first amoeba. What are the chances that we would have said, “I see this going on to give rise to magnificent giant creatures and also puny but smart ones.” To really create intelligence, we must be okay with lack of control. Only then will it produce truly stunning results - mostly not to our liking, but that isn’t the intent of intelligence. Then there is the concept of limitless time. It’s hard to grok what that really means. For instance, if you are in your forties, the distance in time from today to your birth is greater than the distance from your birth to Indian independence. It’s crazy when you put it like that. When it comes to technology, we always underestimate its long-term effects and overestimate its short-term ones. (Amara's Law)

Nature, all of it, is the best example of AGI. But what makes nature tick?

Unplanned evolution: The variety in nature never ceases to amaze. Each organism is distinct from the next, but everything is eventually interconnected in ways we cannot predict. In 1958, Mao Zedong launched a campaign to kill all the sparrows. The rationale was that sparrows feasted on the wheat crop, which led to huge loss of yield. His experts estimated that if they killed a million sparrows, they could save enough grain to feed 60,000 people. But life is seldom so simple. Apart from the grains, sparrows also eat locusts. With the sparrows gone, the locusts had a free run, and within a year China had a famine that killed at least 15 million people.

Self-propagation through reproduction: Nature has both sexual and asexual reproduction systems. Both systems have developed ways in which to mutate from one generation to the next as a way to evolve. However, the evolution doesn’t happen to a plan. Take the example of the British peppered moth. A population of predominantly white moths over a few generations became predominantly black. The change was triggered by the industrial revolution, something in which the moths had no part to play. Even after the trees turned black from soot, it wasn’t like the moths were saying, “Let’s produce black kids.” It’s just that the white ones, now easier to spot, were eaten up by the birds giving them less of a chance to mate and pass on their white genes. 

Time-limitation (systems that die): All life (including the intelligent forms) eventually die. Some die early and others live longer, this is usually the result of either a mutation or a change in the external environment. Generations that die early get a smaller window in which to reproduce, while the ones who live longer get to reproduce more and hand down their mutation to the next generation. For complex systems to work, it is necessary to reduce the number of subsystems to a bare minimum - death. If evolution and reproduction happen without death, life as we know it will come to a standstill since the preponderance of conflicting interests will ensure everything freezes.

Why are these stories important? It explains a deep-seated hubris in humans - we believe we can control outcomes. This will carry into our expedition to build AGI. We will build intelligence with aims to achieve certain goals. But nature, the most complex system we know, avoids this path completely. Things happen and that is it - there is no hidden plan. This allows for evolution to fill a need that may or may not be obvious. For instance, the vulture evolved three times. Each of the three are unrelated. Nature also reuses successful designs. For example, the dolphin which is a mammal, the shark which is a fish, and the Ichthyosaurs which was a reptile all have the same torpedo-like (or hydrodynamic) body structure. Forget being related, they aren’t even from the same species. 

In an AGI world, we would need to build an intelligence that could reproduce itself either asexually (without interacting with another intelligence) or sexually (by interacting with another intelligence) to essentially produce offspring that would pretty much do what the parent(s) do. However, the replicas would have to be inexact copies. Moreover, there would need to be multiple inexact copies so that the most useful at that point in time wins and becomes dominant. 

To expect this hands-off approach from a species that has sculpted nature to meet predefined goals will be asking for too much. From wheat, to dogs, to horses, we have modified everything to be in our service. So we will build AI-like programs, but they will always have an aim.