When we could meet the first intelligent machines

How close are we to living in a world where human-level intelligence is surpassed by machines? Over the course of my career, I have regularly participated in a thought experiment where I try to “think like the computer” to envision a solution to a programming challenge or opportunity. The gap between human reasoning and software code has always been quite obvious.

Then, a few weeks ago, after talking to the LaMDA chatbot for several months, now “former” Google AI engineer Blake Lemoine said he thought LaMDA was sensitive [subscription required]. Two days before the announcement of Lemoine, Pulitzer Prize-winning AI pioneer and cognitive scientist Douglas Hofstadter wrote an article that said: [subscription required] that artificial neural networks (the software technology behind LaMDA) are unaware. He also came to that conclusion after a series of conversations with another powerful AI chatbot called GPT-3. Hofstadter ended the article with the assessment that we are still decades away from machine consciousness.

A few weeks later, Yann LeCun, the chief scientist at Meta’s artificial intelligence (AI) Lab and winner of the 2018 Turing Award, released a paper titled “A path to autonomous machine intelligence.” He shares in the paper an architecture that goes beyond consciousness and sense to envision a path to programming an AI with the ability to reason and plan like humans. Researchers call this artificial general intelligence or AGI.

I think we’re going to look at LeCun’s paper with the same reverence that we reserve today for Alan Turing’s paper. 1936 paper who described the architecture for the modern digital computer. This is why.

Simulate action with a world model

LeCun’s first breakthrough is to figure out a way beyond the limitations of today’s specialized AIs with his concept of a ‘world model’. This is made possible in part by the invention of a hierarchical architecture for predictive models that learn to represent the world at multiple levels of abstraction and over multiple time scales.

With this world model, we can predict possible future states by simulating action sequences. In the paper, he notes, “This can enable analogical reasoning by applying the model configured for one situation to another.”

A configurator module to stimulate new learning

This brings us to the second major innovation in LeCun’s paper. As he notes, “You can envision a ‘generic’ world model for the environment where a small fraction of the parameters are modulated by the configurator for the task at hand.” He leaves open the question of how the configurator learns to break down a complex task into a sequence of sub-goals. But this is actually how the human mind uses analogies.

For example, imagine waking up this morning in a hotel room and having to operate the shower in the room for the first time. Chances are, you quickly broke the task down into a series of subgoals using analogies learned from operating other showers. First determine how to turn on the water using the handle, then confirm which direction to turn the handle to make the water warmer, etc. You can ignore the vast majority of data points in the room to focus on just some that are relevant to those goals.

Once started, all intelligent machine learning tutorial

The third major advancement is the most powerful. LeCun’s architecture runs on a self-supervised learning paradigm. This means the AI ​​can learn itself by watching videos, reading text, interacting with people, processing sensor data, or processing any other input source. Most AIs today must be trained on a diet of specially labeled data prepared by human trainers.

Google’s DeepMind just released a public database produced by their AlphaFold AI. It contains the estimated form of almost all 200 million proteins known to science. Previously, it took researchers 3-5 years to experimentally predict the shape of just “one” protein. DeepMind’s AI trainers and AlphaFold ended up close to 200 million within the same five-year period.

What does it mean if an AI can plan and reason on its own without human trainers? Today’s leading AI technologies – machine learning, robotic process automation, chatbots – are already transforming organizations in industries ranging from pharmaceutical research labs to insurance companies.

When they come, whether in a few decades or a few years, intelligent machines will introduce huge new opportunities as well as surprising new risks.

Brian Mulconrey is SVP at Sureify Labs and a futurist. He lives in Austin, Texas.