Model Mixing Holds Back Artificial General Intelligence (AGI)

Many AI and technology specialists believe that the development of artificial general intelligence (AGI) is only a matter of time. However, there is currently no clear path between existing AI capabilities and AGI. The AGI is nothing more than speculation if there is no traceable path.

Bradley agrees with AI experts who say AGI’s challenge is centered around “learning.” Machine learning (ML) algorithms, on the one hand, learn within the constraints of a single model. The human brain, on the other hand, is a multi-model miracle. Our brain allows us to learn by absorbing and combining various models to build entirely new models. Despite this, model mixing is almost never mentioned in the ML literature. Our brains are constantly combining a variety of models to understand the environment in which we live.

Bradley writes that this miracle of multiple models comes “from the factory” with each of us, and that each human brain lies somewhere between common sense and uncommon insight. His ability to combine multiple models and create new, sometimes unique ones determines his placement.

The Gartner expert believes that we will not be able to develop robots that learn in the same way as humans, unless we break the code of multiple models. “And here’s the kicker: we have no idea how our brains do it,” writes Bradley, who believes there’s a long way to go since billions of brain connections collaborate to absorb, store, retrieve, mix, and build models. .

Abductive reasoning feeds on the mixture of models

Abductive reasoning is discussed in Eric J Larson’s excellent book “The Myth of Artificial Intelligence” (vs. deductive vs. inductive). One of his main claims is that AI as it currently exists is incapable of abductive reasoning. Inductive and deductive reasoning are not the same as abductive. Abductive thinking includes guesswork, brain fog, and even subconscious decisions. The vast majority of decisions we make on a daily basis are based on abductive reasoning. Abductive thinking is based on the ability to combine mental models.

“Have you ever had a brilliant idea that came out of nowhere? That’s what abductive reasoning is all about,” Bradley observes, adding that AGI requires the ability to combine multiple models, something linear language cannot offer. But contrary to popular belief, algorithms do not learn in the same way as humans. There is a significant and fundamental difference. ML algorithms do not mix models. Instead, they “learn” within the confines of a single model.

Bradley stresses that he is not trying to minimize the importance of what has been achieved with machine learning algorithms, as they provide humanity with a plethora of fantastic opportunities.

However, when compared to the learning capacity of the human brain, ML is severely limited. A wide and deep network of various model algorithms would be the ML equivalent of a mixture of models of the human brain.

“We have no idea how to do it,” writes Bradley, who says artificial general intelligence and the possibility of machine intelligence replacing human intelligence should not be discussed by businessmen. Don’t fall for the media hype. Instead, focus on the benefits that machine intelligence can bring to humans.” concludes by pointing out the Gartner analyst.