We welcome a bunch of experts in this interview series. The areas of these people are product management, AI, user experience, usability, big data. Francesco Corea is this month’s guest in our bot enthusiasts series!
He is a complexity scientist and his expertise includes data science, big data, AI, complexity, biopharma, quantitative finance and behavioral economics. Francesco has many publications and awards. Now, he is based in London and Milan. He is passionate about exploiting emerging technologies to solve high-value problems with impactful solutions.
Here is a very enjoyable interview we had with Francesco!
Hi Francesco, since your expertise includes AI, I would like to begin with your insights about it. What kind of challenges do you think AI has nowadays and what will be the challenges in the future? Any maybe solutions?
That’s a brilliant question to start with. Well, I think there are many different problems AI has, but I want to try to focus only on two of them which I consider quite relevant. The first one is a technical challenge. I was investigating time ago how this new wave took off, and I found out looking at trending words (see here for more information) that the first impulse was given by the amazing improvement obtained by Krizhevsky et al. (2012) in the ImageNet Classification competition.
“So, old models used in an innovative way as well as completely new models played a huge role in this AI revolution to come alive. Clearly, nothing would have been possible without (big) data and affordable cloud services, which is the second main factor that pushed the field forward.”
The last feature though is something completely different: I am talking about hardware. We have been able to work on deep learning and other types of AI up to date because we converted GPUs and adapted them to this scope, but the reality is that they are not suitable for the job. If we look five years ahead from now, the spectrum of models used in AI contexts will be open sourced, so models won’t be any longer a barrier.
We are also studying ways to learn more from less data (e.g., one-shot learning, to name only one method), so data will probably not be that relevant (at least not in every industry and/or application). The real bottleneck will be hardware, which requires not only a deep understanding of AI concept and software development but also a clear picture of chip structures. Furthermore, it is important to underline that small companies working in the space won’t probably have a long life because they will either be crushed by big tech companies or acquired fast if of any potential value (Nervana Systems was a good example of that).
The second problem I see is a more general one, i.e., the huge hype around the entire sector. AI is not even close to what media describe and we should be well aware of that because we have already experienced in the past how higher expectations might indeed be detrimental to the whole research and community. There is not enough objective knowledge out there about what AI can do and deliver and it is hard for many people to spot out the difference between facts and speculations.
“It is a very fine line and we need to be careful about it because the system is fragile and a single problem might bring us back 10 years in how we develop, fund and use AI.”
We are all responsible for this hype: media pumping up articles to get more readers; companies trying to find the silver bullet; individuals writing blog posts and renaming their title into “AI-something” to get a salary raise or a new job. It is a vicious circle also because VCs invest into startups and want to get a return out of them, which means either selling to a (usually corporate) buyer or going public. I don’t think the market is going to buy it and I am afraid many corporates will find themselves sitting on a pile of useless tools which might eventually cause a decrease in the entire sector.
How do you think AI will shape our future? It can be a science fiction movie, but as there are some solid things happening right now, how do you project things will change on AI perspective?
It is becoming clearer every day that AI is changing everything around us and in particular, I believe is changing our mindset and point of view. If you are familiar with the movie “Tomorrowland”, I think AI is our “jetpack”. It makes us think that everything is at reach and possible, which is a good thing for the society overall. I also understand that we need to put some serious thinking around the use and design of safe AI instruments, because if from one hand the robot killers scenario is a crazy one and the full unemployment of every human beings on Earth is far to become true, situations where the abuse of the technology might have catastrophic consequences are highly likely (think about autonomous weapons, for example).
“More optimistically instead, I think (and hope) that healthcare will be drastically impacted by AI and boring activities will be assigned to robots and software while we enjoy the best parts of our lives.”
I don’t fully understand yet why we want to build thinking machines or human-like robots, but I am extremely excited about how our relationship with machines is shaping us as humans. In particular, I fully agree with Manuela Veloso’s symbiotic autonomy idea and I deeply believe that we are not simply building better machines over time, but also that they are ‘building better humans’ in turn.
The 37–78 paradigm was indeed thought with this principle in mind: I named this new pattern after the events of March 2016, in which AlphaGo defeated Lee Sedol in the Go game. In the move 37, AlphaGo surprised Lee Sedol with a move that no human would have ever tried or seen coming, and thus it won the second game. Lee Sedol rethought about that game, getting used to that kind of move and building the habit of thinking with a new perspective. He started realizing (and trusting) that the move made by the machine was indeed superb, and in game four he surprised in turn AlphaGo at Move 78 with something that the machine would not expect any human to do.
Hence, intelligent machines will shape the future making us more robotic because they change the way we think and act, but also more humans at the same time because they will free us from things we don’t want to do and allow us to invest our time in activities and people we really care about.
What about conversational AI? How will big data affect on AI? Maybe the analytics side?
Conversational AI is definitely a thing nowadays and we were able to reach incredible progress in a short time frame. However, I think sometimes this is not the preferable type of interface for specific tasks so we should start to look at how integrating it with other tools. I don’t think also that big data will add much more value than it already did in the last five years. We have already achieved a speech recognition at-par accuracy with humans so I believe things will move towards new hardware rather than more data. Voice-specific hardware which optimizes queries responses is probably much more appealing than an infinitesimal increase in model accuracy.
Another problem with conversational AI and Chatbot up to now has been that voice was the lowest hanging fruit when this new revolution began so everyone jumped into that. Of course, many of the applications and companies were a total disappointment (especially because the outcomes were relatively easy to be deceived through expert systems and mechanical Turks).
However, there are finally people out there focusing specifically on pushing conversational AI forward, not only from a research point of view but also from an industry perspective. The Alexa Accelerator (and the related Alexa Fund), as well as Voicecamp (Betaworks), are two players who are actively investing and building new voice-powered solutions in Seattle and New York, but I am ready to bet they won’t be the only ones in a few years time.
In the big picture, how do you think bots will place on people’s daily lives?
If we are talking about virtual assistants or software agents (if we exclude then the physical bots) they are already playing a role, even if we still need to overcome the initial skepticism many consumers might have. I think they will take over many of our menial tasks, although I am not sure this will be immediate. I believe that we should try to privilege more narrow agents which solve specific relatively small problems and only afterwards looking at more general assistants. I agree with Rob May when he said that we will likely have a unique platform/technology to solve many (if not all) of the natural language goals but they reality is that we are not there yet.
If you are familiar with the work of Li Deng from Microsoft, he classifies bots into three major clusters: bots that look for information; bots that look around for information to complete a specific task; and bots with social abilities and skills. I think we are technologically speaking quite far into the first two groups while the third one is still to be fully explored. When this third class will be fully implemented, though, we would find ourselves living in a world where machines communicate among themselves and with humans in the same way. In this world, the bot-to-bot business model will be something ordinary and it is going to be populated by two types of bots: master bots and follower bots.
I have already discussed it elsewhere but in few words, there will be players creating “universal” bots (master bots) which everyone else will use as gateways for their (peripheral) interfaces and applications. The interesting thing about this scenario is that we might be able to partially solve the transparency issue because bots will talk between themselves in plain English rather than sending each other strings of code.
Where are the trends going towards considering big data and AI?
I have spent some time a while ago thinking about what was going to happen next in AI and big data. Many things happened in the last six months which turned out to be not far at all from what I was thinking and observing last year. Technically speaking, we started focusing on nuances of the learning process and I believe this trend will continue and become bigger: how to transfer knowledge; how to remember what it is learnt; how to make an automated decision-maker safe and transparent. These are all questions we need to answer in the process of building something more complete and functional.
There are also two other aspects I would like to underline, which are quantum computing and edge computing. These are two potential solutions for the hardware bottleneck I mentioned earlier. AI is indeed allowing IoT to be designed as a completely decentralized architecture, where even single nodes can do their own analytics. In the classic centralized model, there is a huge problem called server/client paradigm. Every device is identified, authenticated, and connected through cloud servers — that entails an expensive infrastructure.
A decentralized approach to IoT networking or a standardized peer-to-peer architecture can solve this issue, reduce the costs, and prevent a single node failure to break down the entire system. Google was actually working some months ago on a very similar idea called ‘federated learning’, which exploited the computational power of single mobile devices to collaboratively learn a shared prediction model while keeping all the training data on the device itself.
Edge computing, especially in mobile applications and autonomous vehicle, will become essential but quantum computing will probably be completely disruptive. If we might be able to harness the power of quantum physics and break the current limits of the existing hardware we will have access to a new great variety of problems, solutions and applications. A common way to explain the different approach of traditional vs. quantum machine learning is through the phonebook problem.
The traditional approach for looking for a number in a phonebook proceeds through scanning entry by entry in order to find the right match. A basic quantum search algorithm (known as Grover’s algorithm) relies instead on what is called “quantum superposition of states”, which basically analyzes every element at once and determines probabilistically the right answer. There are many things we don’t fully understand yet but the study of the quantum field is fascinating and incredibly powerful.
We will be very pleased to have you as our next guest for this enjoyable interview series. If you’d like to participate in our series, shoot an e-mail to [email protected] !
Don’t forget to sign up for our newsletter to stay in the loop with the latest news!
Source: Chatbots Magazine