Why AI needs political scientists – and vice versa

Your mission is to improve the world's decision-making. How is AI helping you do that?

By allowing me to predict at scale. Governments are the most interesting source of data for AI because they affect everybody. They’re motivated actors who give you every-day primary research on how they've made a decision and what they want you to do in response to it. And they're giving it away free. Using that data and computing power, we have an extraordinary opportunity to improve decision-making.

For example, Bilby can go through everything that a political leader has said about Topic X and put it all into an easy-to-use format. Part one is to make that available to, say, hedge funds or pharmaceutical companies so they can develop their own insights into how their business may be affected, based on everything that a government leader has ever said.

But part two is to hire ex–political science professors, like me, or ex-journalists and other experts, then use AI to plug those additional insights into the data. Our AI gives the customer both the raw data itself, plus our best expert insight. That helps them make better decisions.

That is because decisions are based on data. Every business is becoming a data company, but many don’t recognize it. Governments are also data companies: they’re taking your information and their mandates to rule, and trying to match those two bodies of data, both to stay in power and to make people’s lives better.

But isn’t AI wiping out the marginal advantages conferred by expert opinion? My sliver of expertise can never compete with AI. And, since AI can spew out a huge amount of data, how do you sift the garbage from the gold?

The answer to both those questions is the same: data alone is never enough, because humans are much better at some things than machines will ever be. The two have to work together. So, computers can now beat Magnus Carlsson at chess, but a computer alone can’t beat Magnus Carlsson plus a computer. I doubt that one ever will.

Or take driving. How many billions of dollars are we devoting to train a machine to drive at a level close to, or better than, a human? And yet a 17-year-old essentially picks up the skill of driving after about six weeks of lessons. They may not drive perfectly, but they don’t spend billions of dollars to learn.

We need humans to focus on inference because they’re really good at it. And computers should focus on dumb stuff, because humans aren’t actually that good at dumb stuff. I can’t remember what I had for breakfast 26 years ago, but to a computer, that’s just a data point.

So, decision-making is not just all about data. It’s about using imperfect information, such as bluffs and gambles, and the actions of others that might affect your own actions in unpredictable ways. That lets us decode and use the data.

It’s the intangible motivations that are impossible to input into a program?

Yes. In the same way that happiness is not measured by objective criteria, you really can't define human behavior. But we can break off what machines are better at. And if we can use AI to sift through the evidence for the information you need, you’ll use it more. Then, the more you use it, the better it will work. Whereas, if we’re too reductive, and talk about “computer vs. human,” it’s boring.

Where do you think mobile AI will go next?

Mobile AI will give this power to everyone in their phone. [Clicks on Bilby phone app] I can see everything the Chinese government has said in the last two weeks. But there's 43,628 things.

As things stand today, I can do a targeted search, picking a time period, maybe an industrial sector and a type of news.

But with AI, and Bilby’s offering in particular, I can look, not just at what the Chinese government is saying, but at which sectors are changing the most. I can see where policy is stable, and where changes are thought to be imminent, with the computer assessing the relative influence of who’s saying or reporting each data point. No human can do that.

But this still can't make me money. It just tells me what's changing. The computer gets you to the cusp of where you need to be. Then, the human enters to take the actual decision.

As AI becomes more pervasive and accessible, allowing everyone to tap into that information, what does that do for your business model?

It makes me run harder. Why does Huawei spend more than 20% of its annual revenue on R&D? Because the landscape’s always changing and you want to stay at the top. Can others figure out what I’ve done? Yes. And then I’ll improve.

At the moment, AI is massively increasing competition, yet it’s also massively increasing the returns to existing winners. That then drives my business model: I want to be able to partner with as many firms as possible, and extend my data. I’ve made a cool thing, and I want to get it in everyone's pocket. Huawei has the best kit and great connectivity, so now I want to partner with you. And if I’m the best and you're the best, then no one's going to beat us.

Is there is a moral argument for making connectivity universal?

Yes. Humans everywhere are awesome at finding solutions. We need to give everyone the chance to stick it to the world, because then you're going to find out how good a product is. DeepSeek is the most beautiful recent example of this. As the start-up saying goes: 11-foot wall? 12-foot ladder…

A virtuous circle: get the connectivity right, and you get better AI results and create more capabilities?

Exactly. If you give some random kid in the Australian desert connectivity and a chance, they have an incentive to have a go, because, as the Chinese say “the barefoot don't fear those with shoes.” If you’re from Africa, and someone gives you a chance, you take it or you’ll die trying. And then you get tech kicking in. Connectivity can give you infrastructure, and, if it works, then every time they click that phone, we get more data.

So, it's connectivity writ large. It's also about the plumbing, the boring bricks and mortar, about the people going out with backpacks to fix the pipelines, and so on.

Do you worry that, as AI can sift data ever more effectively, global leaders will evade monitoring by making more decisions verbally, with nothing written down?

A politician who wants his secret machinations to become law will, at some point, need to make those machinations public. And that creates a paper trail of decisions, and of speeches and comments and conversations. At some point, things will leak out, or somebody will talk because they lost out. It's human nature. Again, you can never do everything with the computer.

Do humans have the time, inclination, patience, and language skills to sift all the thousands of documents? Never. So, let the machine do that part. And let humans focus on what they're good at: the networking, the emotional, and the personal. There will never not be old-fashioned human skills – because we’re better. You don't want to go and have a pint with an AI.

It’s the old-fashioned alongside the super-modern?

Yes, and we have to embrace that. Mentality Shift #1 is that connectivity and infrastructure give everyone a chance. Mentality Shift #2 is agency: what do we as humans have agency over? And, if you put incentives in place with agency, and you match incentives to tech, then the circle again becomes virtuous. We need to go beyond the fear and loathing around tech, and get positive.

And “Show, don’t tell.” I can tell you the future is wonderful, but if I show you, then it’s far more persuasive. Today, how often do we talk anymore about our phone not working, or cutting out on the subway? We don’t — because connectivity works, and that shows us that tech makes things better rather than just being told.

It’s about telling a more positive story?

Real success is getting trust and connectivity off the front page because they’re a given, something that people take for granted. And then, once every year or two, there’s a standout moment, such as “What do you mean Huawei have made a bi-fold phone?!” It’s the mix. If you don't have the trust, standout moments don't work.

We need to concentrate more on the human, but we can get standout moments much faster if we work together. It's based on trust and partnerships and humans in the first place. Connectivity is the thing at the heart of it.