Artificial intelligence: The race to the future

Artificial intelligence (AI) has revolutionised the way we live, work, and communicate. From voice assistants to self-driving cars, AI is everywhere. But what is the future of AI, and how will it change our lives?

To answer these questions and more, we sat down with Prof. Dr. Max Pumperla, a data science professor at International Hochschule in Hamburg, and software engineer at Anyscale in San Francisco. He is also the lead author of Learning Ray, a book about the fundamentals of large-scale machine learning, and a contributor to the Ray project, on which models such as ChatGPT have been trained.

1. What is AI & what is AI not?

AI is a subfield of computer science. It aims to create intelligent machines to perform tasks which usually requiring human intelligence. AI systems use algorithms to analyse data, learn from it, and make predictions. AI can be used to perform a wide range of tasks, including image and speech recognition, or natural language processing. We have seen more and more promising applications as of late, and it’s unlikely that the rate of progression will slow down anytime soon. That said, current AI systems still have limitations, as they are generally only able to take decisions derived from the data they are trained on. They often cannot account for all possible scenarios or exceptions, and sometimes exhibit biases or errors. A good mental model is to think of the current generation of AI systems as cognitive automation tools that assist humans and amplify their abilities. They’re not acting autonomously and can’t self-improve or replicate, as some dooms-day prophets might have you believe.

2. The world has been very excited about the new chatbot ChatGPT, with much talk about a breakthrough in the development of artificial intelligence and new possibilities for putting it to use. Did a new digital revolution begin in December 2022?

GPT-3, the base model underlying ChatGPT, will be three years old in June already. The fundamental building block in all such systems, the so-called transformer, has been in use since 2017. In other words, this breakthrough has been long in the making, but it’s certainly worth calling it one. I think it’s a big deal. While models like ChatGPT, or the newly released GPT-4, still have many flaws, even their most pessimistic critics can’t deny one thing: they’re extremely useful. Personally, I use these AI systems every day for coding, writing prose, text summarisation and other tasks. It’s already inconceivable to me to go back to the old times, when I didn’t have such assistance. This is ultimately what technical revolutions always feel like. That same feeling will likely settle in for a larger consumer base, as the products around this technology mature enough. Even if we dropped all AI developments today, I think that the current state of technology would be enough to revolutionise most digital businesses substantially. Recent announcements such as Microsoft’s 365 Copilot give you a glimpse of what’s about to come.

3. What does AI change and how does it change the way we do business?

To paraphrase Amara’s law (American scientist and futurist), we often overstate the impact of technology in the short term and underestimate it in the long run. This is likely the case with the current wave of AI, too. Having said that, it’s not only ChatGPT that should be in the spotlight. We’ve made considerable progress in translating speech to text, generating speech, images or even videos, and many other tasks. For instance, taking a picture of your fridge, asking a voice assistant to tell you which recipes could be cooked from the ingredients you have, and then getting detailed, interactive cooking instructions – that’s not science fiction anymore. Also, the threshold for entering markets will decrease. For instance, with ChatGPT you don’t need to be a programmer to code relatively complex apps. I half-jokingly claim that I was an ok programmer before, but with GPT-based tools I’m an army of one. In day-to-day business, expect these AI systems to help you ideate and write good first drafts of all sorts of reports, transcribe and summarise meetings, give you complete suggestions for email replies, help you do analytics in your Excel sheets or create data visualisations. In other words, a lot of the grunt work will be automated, and we’re free to focus on more strategic topics. For me, it’s already a game changer to be freed from so many menial tasks, and I don’t think anyone will miss setting up a PowerPoint presentation. Beyond that, it’s difficult to judge the long-term impact at this point. 2023 is going to be an interesting year.

4. There is a lot of hype around AI, to the point that AI will be driving our cars, cure cancer and take over all our jobs, and it’s difficult to separate hype from reality. Where do you think AI is now?

On a high level, I find it useful to consider Daniel Kahnemann’s two types of thinking – fast and slow . Humans are good at both quick thinking, such as visual or auditory perception (e.g., if you show me a picture of a cat, I immediately know it’s a cat), and slower, more methodical thinking (e.g., contemplating where exactly the cat might have gone since breakfast). Current AIs are very good at fast-thinking tasks, and while we do have excellent, specialised AI systems that can solve slow-thinking tasks such as playing chess or go, the recent wave of so-called generative AI usually struggles when you want them to be precise. Therefore, when accuracy matters, we often can’t defer to AI systems. To give you an example: If a pedestrian walking from left to right “vanishes” behind a small bicycle shed, a human driver observing this situation might reason that this pedestrian will show up to the right of that shed in a couple of seconds. But the pedestrian might also stop or turn around again. Human operators know to be cautious in these situations, but AI systems tend to struggle with them, for various reasons. Autonomous driving is a challenge, and this is just one example of why we have been seeing it get pushed to “next year” for years now. At the same time, there are indications that the latest generation of GPT-type models seems to understand “theory of mind” reasoning, the type of he-knows-what-she-thinks-what-they-think thought processes that we thought was exclusive to conscious beings like us. If that turns out to be true, which I’m still highly sceptical of, this could be a step towards many of the harder problems we face. Recent studies seem to suggest that it might in fact be the jobs of highly qualified workers that fall first. But even if many jobs get displaced faster than we thought, I’m optimistic that we can find new and more interesting things to do, as humans always have when faced with a revolution. I’m convinced we can make this a net positive for everybody.

...we often overstate the impact of technology in the short term and underestimate it in the long run.”

5. Are there any sectors particularly well placed to benefit from AI?

Any sufficiently digitised industry sector can benefit from recent AI advances. Instead of thinking about sectors, it might be better to think about the type of tasks to automate. Depending on what you want to do, there might be limiting factors such as data availability or security. For instance, you already find companies today detecting cancer from digital images, but those images are highly proprietary and private, which makes these  types of companies hard to build. If you have access to high-quality data and clearly defined tasks that don’t require too much interaction with the analogue world, chances are that you can automate at least parts of your business, today.

6. How do you think we will look back on this time in 30 years’ time?

In my mind, there are two scenarios. Either we look back at this era of modern machine learning and view it as the point in time when AI took off and we never looked back. The other option is that the current techniques, which mostly rely on using ever more data and compute power, and not so much fundamental mathematical breakthroughs, will hit a wall. My take is that we’re still missing at least one crucial step before we can truly reach human-level intelligence. I hope we get there in our lifetimes. In either case, I think it’s likely that what’s happening right now will be mentioned in the same vein as the internet – the time when AIs finally became  useful for everyone.

Interviewed by: Laura Kuenlen, Investment Communications, 21 March 2023