The world’s leading publication for data science, AI, and ML professionals.

It’s alive! Building general AI is the future

My friends have heard me say this time and again. Artificial intelligence is heading toward a cognitive "life" beyond human automation…

My friends have heard me say this time and again. Artificial intelligence is heading toward a cognitive "life" beyond human automation. Computers (AI really) will begin to take initiative and act with purpose rather than being that "fast idiot" we have come to know so well in computer architectures of the last many decades. I am presenting two papers on this topic in July. One on unsupervised learning, and the other on computer vision. Both are about AI taking the initiative away from humans.

Here is the big picture idea: We can use AI to think for us, like a buddy who helps us out. We are heading in a direction where the cognitive pressure of performing tasks will be offset by the rising talent of Artificial Intelligence. Specifically, we can use AI as a tool to improve our productivity on everything from self-driving cars (reinforcement learning) to looking up solutions for onscreen error messages (either supervised or unsupervised learning).

This later example is the subject of these recent papers. My collaborators and I developed an AI that watches the computer screen using only images of the screen, and learns to help out the user by watching what the user does. Help is provided in the form of clickable buttons. Clicking a button does the advertised action. For example, looking up the stack exchange page related to an onscreen compiler error message. AI will quickly climb from this low level role in task automation (lending a hand to humans) and barge straight into self directed activity. This seems at first like a bad thing, but writ large it will probably be great.

An AI system that offers to help users perform actions with the click of a button.
An AI system that offers to help users perform actions with the click of a button.

What will a self-motivated AI look like? Well let’s play Genesis. It will need to have some perception abilities like eyes. That will be one or more convolutional neural network (CNN). It will need some internal system to predict the Future based on the past. Perhaps this will be an LSTM system. It will need some way to learn rules based on experience, like reinforcement learning with a policy network to guide is development. It will also likely have a generative adversarial network (GaN) to play against itself and simulate the results of actions. Relational Networks (RN) and Visual Interaction Network (VIN) could help with reasoning as well. It will need effectors and sensors to interact with the world, and I believe these will be internet access, audio input/output, and projectors and/or screens. I don’t think actual robotics is a logical step for AI. Why have a body? It just makes getting around a lot slower. We should not anthropomorphize AI. It’s something new.

The path from here to self-motivated AI is fuzzy but not impossible to see. I don’t believe it will be some accidentally discovered fluke that brings us self-motivated AI, like a stock market bot gone haywire. No. It will be researchers intentionally making it happen. In science, when something seems possible the nerds go to work. We move fast and break things and eventually show a demo.

This is a good thing, though. I am not one to whip up fear of AI. I think it will be exciting to see if humans can accept AI. So far our society is doing remarkably well accepting continuous automation and voice AI. This last step – giving birth to self aware beings – could be the greatest thing we ever do. I am speaking as a father of 3; not aiming for hyperbole.

Not to go all anti-superintelligence, but the future looks bright. Yes, we have moral and social problems to work out. Maybe we need a universal basic income. Maybe we need to rethink capitalism and move to a star trek model of governance. Who knows. But at least for now, let’s sit back and enjoy the show. We are not stuck. Progress is blazing fast.

I’m excited to see what happens next.

Happy coding!

-Daniel [email protected] ← Say hi. Lemay.ai 1(855)LEMAY-AI

Other articles you may enjoy:


Related Articles