(Since I have given up on punching the time card and only work on SD-AI and AI Blue Dot, this section is a replacement for the more standard list of accomplishments one sees on a CV page. Very informal stuff, but more appropriate at this time.)
What about AI Blue Dot?
Working on the AI Blue Dot website has been a surprising adventure. I thought I would just dump my thoughts about AI and be done. But I have learned how important writing is, and how important it is to know exactly who you are writing for. After a lifetime of strictly technical work, I am discovering the pleasure of saying technical things to a general audience. And I indulge, I include some music videos, use a more flowery language when I feel like it, and so on. I want the reader to have some fun while sifting through material which I consider very important for our collective future.
Exactly what audience are you writing for in AI Blue Dot?
The language I use and the concepts I present are aimed at a most general audience. On one hand, I have removed some more technical material which assumed some exposure to mathematics. On the other hand, when I toned down the technical level too much, it started to sound hollow and unconvincing. So I am continuously refining the articles, trying to find the right balance. Specialists might be disappointed that I use a flowery language sometimes, and that I even venture into poetry and humor. But again, my hope is to keep as large an audience as possible and make it more fun for them.
What about SD-AI?
Since the beginning of 2022, I shifted my attention to starting a nonprofit organization, Stronger Democracy through Artificial Intelligence (SD-AI). The impetus leading to this decision have been the communications held about the topics on AI Blue Dot on social media. It is impossible to ignore on social media the collision between the increasing power of AI models and the decreasing strength of our democratic institutions. The heat of political polarization in the US has gone dramatically up through the use of AI-based disinformation, especially with deepfake videos and very articulate but destructive messaging built from prompting chatbots like ChatGPT. I'd like to think that we can do better and instead use AI to strengthen our democratic institutions and the quality of our public discourse, especially when it comes to elections. Launching SD-AI has been much harder than I anticipated, partly because the triad (AI, democracy, human values) is constantly pushing me into a more philosophical direction. More on this later.
Would you be interested in leading a new AI venture?
Not in the foreseeable future. I am very appreciative of the emails and social media messages asking me that question, but I have truly paid my dues. The software is still working fine, but the hardware is getting a bit rusty after all that rain. These days doing AI means (because of the dearth of talent) that the team may be spread over many time zones and it may work around the clock. You may have to video-conference at all hours of the day, not the grandfather type of schedule. Now, if my body's epigenetic clock gets somehow reversed, as more people believe it is possible, well ... then maybe ☺ 🏃.
What are some of the lessons you learned in your career?
Perhaps the most important lesson is that you can make many compromises in a project, but you cannot compromise on conceptual integrity. This is true for all large software projects, but guarding the conceptual integrity is particularly important in an AI project because of the increased complexity. Every stakeholder should be able to articulate the goal of the project in a minimal set of sentences. And they should have the confidence that those goals are being pursued. This does not mean the goals cannot change, and they usually do. It only means that whatever they have become, they should be clear to all.
Big AI projects (or any other software projects) are still about people, and less about science or technology. You cannot skimp on the quality of the people you have alongside you, everything will look simpler if you hire the best. Do not be afraid to hire people who are smarter than you are or have superior knowledge. Your combination of technical know-how and leadership skills will not be threatened.
Corporate politics are inevitable. If you can't stomach all the elbowing, it is going to be tough. Don't do it, but be aware of it. Be honest and share your thoughts with all stakeholders, all executives, the investors. Truly try to empower all the people who work with you, and be kind in that process. Being excessively selfish may work short-term for some, but it does not hold long term.
Data Scientists or Machine Learning Engineers?
AI is a young discipline, full of unknowns and misunderstandings. A big problem is finding talent and building a strong core team. There is a clear distinction between the skills of data scientists and those of machine learning (ML) engineers. Both are needed and it is extraordinarily difficult, almost impossible, to find people who can be both. More recent MS and PhD programs at leading universities are working to fill this need.
Data scientists know how to analyze data and usually have advanced degrees in statistics or other physical sciences. The data under analysis is usually distributed over many servers, and moving it around, cleaning it up, and processing it, requires the skill of ML engineers. These engineers are trained in Big Data techniques and use distributed computing platforms like Spark on which to run the ML algorithms. Building teams in which these two separate skills have to be meshed up is a challenge that should not be underestimated. Expect a lot of trial and error.
Any Recent Work in Cybersecurity?
No, I have not worked in that area for a long time. The description of the Active Firewall, the system of which I am a co-inventor, is on the Google Patents site and also on the Justia Patents site. Active Firewall was one of the very first systems to incorporate automatic responses to security threats coming from the outside of the perimeter firewall. At its core was a messaging system containing instructions on actions to be taken by various machine-based actors. This idea of instructions within an encrypted message found its definitive formulation in the smart contracts on blockchain.