”The main thing with AI is that it can handle a lot of data and complexity that we can not handle ourselves”
On the current and future state of AI
October 12, 2022
Who are you?
— I have a long academic background with triple degrees, a Master’s in Mathematics, a Master’s in Engineering Physics, and a PhD in Engineering Science. I was one of the inventors of the 5G communication system. Since I quit my job at Ericsson, I have worked part-time as a researcher within AI and Machine Learning and part-time as an entrepreneur, founding three companies. I’m now mainly working with my company OrganAi.se, a digital assistant helping people organise their daily life. Typically, if you want to meet someone or several people, you need to text or e-mail back and forth to find a time. What we do is that we synchronise the calendars and automatically find the time, so it takes less than ten seconds. And I’m also doing some research and consultancy, consulting people within AI — from the strategy process and the product, but also very deep technical.
”In Europe, we’re about to lose the race when it comes to AI”
If we take the broader perspective, where are we now in AI? How far have we come?
— There are some companies that are making great progress and even progressing in the state of AI technology, and that includes big companies like Google, Apple, Amazon, Microsoft, and IBM. They are making even more progress than universities — at least when it comes to applications. I’m not saying that there is no interesting research coming out from academia but we really see that the revolution is coming from companies nowadays.
— And we also see progress in art, most recently with DALL-E, where you can generate art with the help of AI — both beautiful, abstract patterns, but also painting-like pictures. It’s the entire spectrum, says Gattami. He continues:
— We don’t have the kind of intelligence that humans really have. On the other hand, we already have intelligence that can beat human intelligence. It would be cool to have intelligence like human beings but I’m not sure if it’s really the goal. I think, what’s important is that we have technology that helps the human race forward, for our survival, and helps to solve problems that we can not solve ourselves. The main thing with AI is that it can handle a lot of data and a lot of complexity that we can not handle ourselves, and I think that this is what we should aim at. That said, it would be really nice from the user experience point of view, in many products and similar, where we have an interface that is actually some kind of chatbot and we communicate with devices in that way instead of pressing buttons and trying to find ourselves around in software or so.
— But there’s also a huge difference and it has become a large gap between companies. To be honest, I think in Europe, we’re about to lose the race when it comes to AI. In the UK, they have DeepMind, which is great, but that has been acquired by Google. So, whatever happens in Europe is related to companies and their research departments that have their headquarters in the U.S., such as Google, Amazon, Apple, and Microsoft. When it comes to pure European players, I can not come up with one single company that has some kind of leadership in the area whatsoever. Maybe Volvo is getting somewhere with autonomous driving and I know that they’re betting on it but, as we know, the problems with the technology are not solved yet. So, we’re lagging behind here, while in China, there is a lot of progress — it’s basically in the US and China where I see it happening.
You mentioned cars and we see AI in beauty tech, consumer goods, and many other industries. When you’re contacted by companies, what are they looking for?
— Since we’re lagging behind, many companies don’t really know what they want and we (the labour market, Ed’s note) don’t have enough expertise to serve them. If I look at Sweden where I’m based, it’s typically a culture of your ’network’, so you just have someone that has been working in the company for a few years and then suddenly becomes responsible for the department of AI, or data science, or whatever they call it. And this person might be brilliant but then, he or she doesn’t have the background required to run these things. So, managers aren’t really looking for the right thing and that’s why we don’t see many things happening.
— I think we need to change our approach. The gap is huge — the salaries in the US are not twice, not three times more but in many cases four times more. It’s clear that we haven’t realized here how we should do it. And also, companies that really want to be AI-driven, need to be data-driven and bet on how they should collect information and data. With this data, they can do so much stuff, but that also requires a lot of different thinking. You really need to think about the user experience, and how easy it is to insert data for the consumer and for the consumer to share their data in order to do something for them. As mentioned, the huge gap is not only about technology but also the way of taking and approaching the whole thing.
And where to start and what to focus on for ventures when it comes to AI?
— You need to have the right competence in place and that’s where you should start. They don’t need to understand the technology but what you can do. Maybe look at success stories from other companies, how they collect the data and think about the product, the user experience, and the whole thing, and then hire the right people and work with experts who also know the technology. Ask yourself: What can we do? What can we not do? What are the low-hanging fruits? You need to have a road map strategy and understand how you build these things. So, you need to have the right team, which means that you need to have people that have experience in the business, understand what data you have and what you could collect, and the value of it — what you can get from collecting it.
Can you share any predictions on where we are with AI in three or five years?
— I hope that we’ll have more efficient ways of training these AI models. Now, it’s data-hungry and training them consumes a lot of energy, which is not sustainable. And I hope that we’re gonna have methods that we know will do their best every time. Now, it’s random and you don’t know — sometimes they perform bad, sometimes they perform well, says Gattami, continuing,
— From a more application point of view, I hope that we’ll have AI systems that can do some more reasoning. Now, it’s a lot of pet recognition and predictions. For instance, if you apply AI in healthcare, it says: ’You need to have heart surgery.’ It needs to explain why it’s necessary and you need to understand, so it needs to do some reasoning and we’re not there. It’s not like we can ask: ’Why? What are the consequences? How do they see that from all the data?’ I don’t bet that we’ll ever have this complete smart system but some would be great. I think we maybe start to see the human-computer interface — devices with any kind of computation like cell phones — where the interface is maybe more conversational rather than pressing buttons and that’s something I can see in five years.
— We’ll maybe see autonomous driving but I would say closer to 2030 than 2027 or 2028. We’re getting closer though, and we’ve actually come very far. The problem is not autonomous cars themselves, it’s humans… Humans are the challenge — if we were to eliminate all human drivers, we can do it today. Unfortunately, you can’t.
— I also think that AI will help us in trying to build smarter methods based on smart data sets than the ones we need today. If you look at the human being, I can take one example from Yann LeCun, who’s a very famous AI researcher. He said: ’look at this 4-year-old kid, they learn all the time and get smarter, more and more intelligent — and this is how we would like the AI systems to be’. So, you have something that you don’t need to teach from scratch but basically more of a continuously learning machine. It’s not from scratch, it should be something that builds and learn more things and more things and more things. This is called ’Transfer Learning’ where you have many examples and you’ll be able to mimic the same things, based on small data. You’ve already learned much similar stuff, so to do the new thing, you don’t need to have a lot of data for it.
— Yes, so you really need to think about new approaches. Some things work well but how can we do them even better? It’s a lot about learning — how we humans learn, basically, says Gattami.
Gattami’s key takeaways:
1. The world’s biggest ventures — Google, Apple, Amazon, Microsoft, and IBM — and not the universities are making the most progress within AI.
2. Europe is lagging behind — it’s in the US and China that the biggest AI development is happening.
3. The companies that want to be AI-driven need to be data-driven and bet on how they should collect information and data.
4. We don’t yet have the kind of intelligence that humans have but on the other hand, we already have AI intelligence that can beat human intelligence.
5. The next step can be AI systems that can do some more reasoning.
6. A computer interface that will be more conversational rather than having the user press buttons.