I have been out from the AI conference circuit for a while now, mainly because I find many of them quite repetitive and not very informative. Do not get me wrong, there are many good conferences out there, but very few where the content is placed before companies exhibitions or sales pitches.
And if there is one event in Continental Europe where “the content is king”, that is definitely the Rise of AI conference.
A one-day event, gathering 600 hundred participants in a wonderful location in Berlin, and dealing with over 40 talks over three major tracks, all greatly organized and managed by Fabian and Veronika.
I happened to learn a thing or two at the event, which is what I am going to share below.
I. AI and Blockchain are a real deal
You might know by now that I am really keen to see and study how those two technologies interact (and when you see Ben Goertzel on stage, everything gets a completely new sense, so expect some new thoughts soon).
In my view, one can have positive effects on the other and vice-versa, and more specifically AI can make blockchain more energy efficient, scalable, secure and potentially useful (more details and more points here), while blockchain can actually make AI more explainable, trustworthy, effective, secure and democratized (again, for more details look here).
As I already previously stated, blockchain and AI are the two extreme sides of the technology spectrum: one fostering centralized intelligence on close data platforms, and the other promoting decentralized applications in an open-data environment. However, if we find an intelligent way to make them working together, the total positive externalities could be amplified in a blink.
II. This AI technology wave is different
This point embeds different nuances: first of all, it is still not clear to everyone how to implement an AI solution within an enterprise context. Working with AI means continuously experimenting, so pick a use case, start building, testing and iterating as quick as possible rather than looking for the perfect mathematical (academic) solutions (have a look at the OODA loop, i.e., observe-orient-decide-act), and only then scale and impose a transformational shift to your organization.
Second, I have spoken and listened to investors and entrepreneurs all day long, and basically, no one mentioned IP protection of any kind. I have been thinking about this issue for startups for some time now and I keep finding evidence that patenting innovation belongs to the old-world of innovation while open-source technologies dominate the AI scene (unless you are a very early-stage company — more on that here and here).
Finally, building AI simply for the sake of doing it is a waste of time, while using it for solving a real-world problem is extremely powerful (and well-seen by investors). Using Christian Nagel’s words (a partner at Earlybird Venture Capital), you need to move from a pure data-centric approach to an action-centric one and invest in team, category leadership, disruptive data acquisition strategies and business models (e.g., integrating models into customer processes or 3rd party systems, building infrastructure or platform to allow the construction of models and the value exchange between the parties, etc.).
III. AI is not all peaches and dandelions
There is a sane skepticism about how press depicts AI and the future trajectory the field might take. I have heard the same thing over and over: machines do not understand (i.e., the Chinese room argument), they do not have a human-like brain (and likely never will), and superintelligence is more a horror story to be sure your sons well-behaved rather than a possible short-term scenario.
However, that does not rule out the possibility of having in the future a general-purpose system of algorithms supporting people in their daily activities, but it asks for a good dose of realism and a renovated approach to solving the issues posed by stacking up several layers of non-easily-integrable technologies.
In this fashion, Peter Bentley suggested that in order to make a Strong AI, i) we should not try to emulate human intelligence since true intelligence emerges through the need to solve complicated problems and it is general and adaptable; ii) there is still a lot of uncertainty of which approach or algorithm will pay off, and you need to consciously build new layers of the stack without losing the functionalities of the previous ones; iii) testing is never enough, and increasing complexity demands for an increasing number of tests to build a safe AI.
With these principles in mind, you can try to overcome three main challenges to achieve a general AI, as pointed out by Chris Boos: the transfer of knowledge (and the catastrophic forgetting problem), the Moravec’s paradox (high-level reasoning requires little computation, while the opposite is true for low-level sensorimotor skills) and working with small datasets (one or zero-shot learning). A few attempts exist to solve some of those problems, but much more research is still required to make them functionally attractive and scientifically accurate.
IV.Germany is still a big AI player
Technical research is very strong in Germany. Whether this actually translates into an operating business model is though a completely different story, but there are enough data points in the startups’ ecosystem to qualify Germany as one of the countries driving the field (mainly together with UK and France).
From a research standpoint, institutes like the German Research Center for AI, the Max-Planck Institutes, the Fraunhofer Institute or the Cyber Valley are leading institutions studying AI from the ground up (and the government itself has been highly supportive to numerous AI projects in the past). If you add up to that a hundred early-stage AI companies, a strong industrial ecosystem nurturing and investing in AI applications and a few strong venture investors, you have the right ingredients to have a very tasty meal.
Fabian has more thoughts and data on this topic here if you want to check the German ecosystem. He also outlined the challenges and opportunities he encountered in the last 4–5 years investing in the space. In his own words:
- It is difficult to understand and evaluate Artificial Intelligence investments;
- There is a lot of hype and a lack of substance with many teams;
- Funding for AI startups is tougher and the seed stage takes longer;
- The market potential for applied AI is huge;
- Teams often have a high level of education and research background but lack entrepreneurial experience;
- Once AI works, it is hard to replace.
In addition to that, don’t forget to download his just released Global AI report here!
Well, I really hope you enjoyed the conference as I did and that it was inspiring and provocative as it was for me.
I hope to see you next year in Berlin!