GenAI a Phase: Mustafa Suleyman
Just when you thought it’s here to stay, the DeepMind co-founder says Interactive AI is next
First there’s the shock and awe! Then there’s the realization that there’s only so much that next gen innovation can achieve. Which then leads to creative minds taking over artificial intelligence and making it more intelligent and less artificial. Well, we just made this up with a chatbot’s help – thanks to a thought-provoking idea shared by DeepMind co-founder Mustafa Suleyman.
Suleyman had a lengthy conversation with Will Douglas Heaven in the MIT Technology Review where he articulated that generative AI was just a phase and would soon be replaced by what he termed Interactive AI. One where bots run tasks set by humans through support from other software and people to get things done.
Move over AI, Suleman says Pi is here
His latest venture Inflection.ai, is already a billion-dollar company, and boasts of top-tier talent from companies such as OpenAI, Meta and DeepMind. In May this year, the company introduced Pi, a personal AI assistant that offers conversations, friendly advice and concise information in a natural and conversational style.
Pi, which stands for personal intelligence, is not just smart but has good EQ, Suleyman had said after the launch. Now, he confirms that the efforts his company took to make Pi controllable is yielding results. You cannot get Pi to produce racist, homophobic, sexist stuff. Nor can you get it to coach you to produce a biological or chemical weapon, says Suleyman.
Without going into the details of how a chatbot achieved this degree of EQ, Suleyman says for years he has lived with the idea of espousing human values to make the world a better place. So, if you start complaining about immigrants in your community taking jobs, Pi will not call you out or wag a finger. It will inquire and be supportive to understand where it comes from, while encouraging the user to be more empathetic.
The future of AI isn’t limited to chatbots
Referring to the growth of AI, Suleyman says the first wave was about classification that showed a computer to be trainable to classify various types of data such as images, video, audio and language. The second phase now relates to where one takes such data and produces new data through a modeling process that teaches and learns.
The next will be an interactive phase where conversation would be the future interface, Suleyman says talking to your AI assistant will be how we communicate. And these AI tools will be capable of action whereby a high-level goal set by you could be implemented by it via conversations with other people and other AIs.
Suleyman thinks that one cannot underestimate how technology is going to grow, from a static one today that just talks to going animated and with the freedom to take action, if granted by the user. And this is where he believes that humans will always be in command through setting up boundaries that AI cannot cross.
And regulation isn’t going to be too difficult
He suggests that creators of such technology can actually set boundaries of safety right from the actual code to ways it interacts with other AIs and humans to the point where it can even limit the motivations and incentives of companies that seek to create such technologies. Suleyman thinks that if humans use technology for improving society, things can get easier.
Which brings us to how he perceives regulations at a global level. While Suleyman is happy with the efforts for creating oversight, he believes boundaries must operate at a micro level too that the creators must set right up front. So, creators must be wary of certain capabilities right up front and not rule them out as potential mischief points.
As examples, he notes that the AI should not be permitted to go and update its own code without human oversight and government agencies can even consider licensing this activity, just as they do when it comes to nuclear material. Suleyman is quite convinced that regulating AI wouldn’t be as big a problem as it’s being made out to be.
He calls the current state of affairs akin to a panic attack when one is faced with stuff that one cannot adequately comprehend in the first place. He points to the limitations placed on drone usage or even internet privacy initiatives as an example that the industry would get together and create guidelines and guard rails for developing artificial intelligence.
