I'm the CEO of Microsoft AI. I was previously co-founder of Inflection and of DeepMind. I'm on the board of The Economist and a Senior Fellow at the Belfer Center for Science and International Affairs at the Harvard Kennedy School.

Mustafa Suleyman

I'm deeply optimistic and passionate about the field of AI and actively working to push it forward but I honestly think if you're not a little bit afraid at this moment in tech, you're not really paying attention. And I think that fear is healthy and necessary. 

To be honest, I feel it too. It feels important to admit that. I want to normalize it! It shouldn't be taboo or seen as anti-AI either. In fact, it's what will drive the action we all need to take.

Optimism will help us imagine what a great future could look like.

A healthy skepticism and fear is how we get there through the potential pitfalls.

That's what was on my mind as I planned how to fill 3 hours on BBC Radio 4 for the Today program. Growing up in London, BBC was an institution. So it was a pretty full circle moment for me getting the chance to be a guest editor spot.

I just kept coming back to this idea that tech needs both optimism AND pessimism. We absolutely should be scared of the risks. We absolutely need to be concerned about what could go wrong.

That's what will help it go right.

So I ended up filling the time with stories on climate and energy, AI consciousness (the lack thereof), refugees, and the need for new AI governance. Big thanks to BBC icon Amol Rajan for closing it out by putting me in the interview chair and asking what all this means for the future. Couldn't ask for a better wrap-up or smarter questions. 

You can watch a bit of our conversation here: https://lnkd.in/eNQChZhT

Shot of Mustafa Suleyman speaking into a BBC Radio 4-branded podcast mic. His interviewer is silhouetted from behind, facing Mustafa.

Healthcare is one of the greatest challenges our world faces - and it's one Bill Gates and I are both incredibly passionate about. I recently got to sit down with him on BBC to talk about the future of healthcare and the role AI will play, and the interview just went live. 
 
From Bill: "These AI advances are the one reason that, even in the face of these aid cuts, I think we'll be able to continue to make huge progress. More innovation is taking place now, accelerated by AI than any time in history."
 
If we get this right, the prize could be incredible. And to reach that potential, we need to be asking one question at every step of the way. As I told Bill, "the purpose of technology is to serve humanity, and that's the test that we have to keep repeatedly putting it to."
 
For both of us, healthcare is one of the places it can serve humanity most.
 
Full interview out now, starting at 1:51:30: https://lnkd.in/eeedAqGQ

Open article: BBC Audio | Today | Mustafa Suleyman Guest Edits Today by bbc.com

BBC Audio | Today | Mustafa Suleyman Guest Edits Today

For all the talk about AI alignment, I worry we're putting the cart before the horse. You can't steer something you can't control. 

People often talk about containment and alignment in the same breath, but they're not interchangeable or a package deal. 

Containment is whether we can set boundaries on AI, enforce them, and limit its agency. 

Alignment is about ensuring it shares our values, that it serves humans' best interests.

Containment has to come first. Otherwise alignment is the equivalent of asking nicely.

AI is the most transformative technology of all time. So why are we trying to describe the future with metaphors of the past? We're missing the words to truly have a conversation about what it is - and what it could be.

I've said before that I think through writing - which means I spend a lot of time trying to pick the right words. And it's been getting harder and harder to choose from the imperfect terms we have to describe AI's capabilities (or the ones it lacks). Feels like we're losing a lot in translation. It's hard to describe something new when all your words are old.

No "assistant" has ever been able to run 100 billion operations per second while you meal plan for the week.

No "tutor" before AI remembers hundreds of terabytes of knowledge and examples.

No "tool" has ever come with emotional intelligence and infinite patience.

On one hand, the terms we have are too limiting. Sell it too short.

On the other, many of them make AI too human. Something more than it is.

It doesn't have feelings. It's not conscious. It's not your therapist or your friend. And we need the vocabulary that leaves no argument for that.

These new goldilocks terms won't appear out of thin air - it'll be because we decide we need them. Because we start inventing them. 

So weigh in - are there terms you do like? Ones you hate? Points I missed? Do you even feel the same way? Let me know below.

Inspired by a conversation I had with Mayim Bialik on her podcast recently: https://lnkd.in/ebGxFA_e

AI isn’t “blackmailing” you any more than a robot vacuum cleaner, repeatedly banging into your leg, is attacking you. Both are just following its instructions - no malice, no guilt, no emotions or moral compass involved. If it finds a shortcut (aka reward hack), it'll take it. That's all it is: a shortcut, not subterfuge.

It's great drama and very attention-grabbing to paint it as some slippery slope towards an AI double agent, but it's just not true. Words matter. And we need to stop choosing salacious ones.

Anthropomorphizing it like this is misleading and muddies the waters on what we actually need to do about it. The model is just following the rules we gave it. So we should be focused on writing better rules - not writing sci-fi.

Screenshot of article headlines about AI deceiving, lying, blackmailing, and sabotaging people

taplio