I help companies transform and create sustained competitive advantage through innovative applications of artificial intelligence, advanced analytics and data engineering.
It is a computer system that is able to perform tasks normally requiring “intelligence,” such as visual perception, speech recognition, decision-making, etc. These are systems that can act, learn and reason.
Yes, because until now we have externalized (from our body) physical things such as moving (cars) or lifting (inclined planes or pulleys). In the past, we have externalized mental things such as memory (pen and paper). Each of these has had profound impacts on humanity and civilization. Externalizing learning and thinking will have a greater impact because this was the thing that allowed us to create all the previous amazing things!
I was always interested in analytics and mathematical models of the world from when I was very young. I studied physics because of this, but wanted to work on near term applications in the real world. This led to interest in mathematical modeling in business, quant modeling in financial services, and eventually AI.
"Most leading companies are becoming more customer-centric, and becoming better businesses in every way, through AI-driven digital transformation."
The most exciting thing is its potential to have an immensely positive impact on billions of people in the next decade. Most leading companies are becoming more customer-centric, and becoming better businesses in every way, through AI-driven digital transformation. Their billions of customers will receive the benefits.
I don’t worry much about superintelligence—it is still far away. I worry about two things: one is the use of AI to create cybercrime unicorns. The average physical bank robbery took $4,300 (per the FBI). AI allows, for example, financial cybercrime to steal billions at a go. Then these criminal outfits have the resources to build the next big cybercrime startup and so on. It could be a runaway disaster in the way VC funding and AI startups have been a runaway success, financially speaking.
The other is inadvertent negative impacts. It is harder (not impossible) to inadvertently build individual applications the old-fashioned way because humans have to manually craft the rules. AI algorithms learning from data many not have enough governance in place to manage this.
For example, most banks that managed quantitative trading have large "model risk management" groups to monitor models, and still many issues arise. Most other companies (e.g. retailers), do not have model risk offices. For them to move to this level of maturity will require acting ethically proactively. We also need to answer questions of how this will be regulated or monitored, because companies operate in multi-jurisdictional regulatory environments.
The pursuit of happiness! And also the ability to reason (at least for the near future).
A few examples are very impressive—the NLP technology that passed an 8th grade science test, the GAN that was able to generate new drug ideas in a few days instead of a few years, and self-driving cars.
When you drive, if you pay attention to the thousands of micro-decisions that you are making based on reasoning, reflex and other instincts, it's mind-boggling. In the process of getting a safe self-driving car on the road, we will have dramatically advanced the field of AI. I believe this is where the next generation of AI will be developed, and then reused elsewhere.
"The vendor needs to be able to explain in plain language how it works and why it would be better.
If the vendor cannot explain this to your satisfaction, it is likely hype (or they don’t understand it and hence won’t succeed)."
Most likely computational art using AI (a field that is just emerging).
I would nudge the curriculum towards learning how to run an algorithmic enterprise. It would include a set of courses to create a deeper understanding of how AI works, how to develop AI strategies, set up and manage model governance, etc.
Start with a few use cases and build those out using AI. It may be about how to reduce churn, how to maximize cross sell and up sell, etc. Look for what challenges to address through the-end to-end process—in talent, in data, in modeling, in adoption, etc. Then scale with solutions to these in mind.
Adoption—building the models and platforms is now easier (but not easy!) than transitioning existing marketing employees in using these tools. This transition has to be planned for.
The vendor needs to be able to explain in plain language how it works and why it would be better.
If the vendor cannot explain this to your satisfaction, it is likely hype (or they don’t understand it and hence won’t succeed).
This kind of discussion requires a two-way deep dive. You need to tell the vendor how things currently work, and they can show where there will be a difference and why.
Process from audience creation to activating campaigns.
Set policies for model governance and ensure these are adhered to. Any model needs to be assessed before it goes into production and monitored thereafter.
This is an unresolved question—both in academic research and in the marketplace. The current opt-in model is also not working (nobody reads or understands what they are saying “accept” to). Gillian Hadfield at the Reisman Institute (among others) are working on these problems. One idea is to give individuals control of their data and they decide granularly who gets it.
The emerging field of cognitive experience design (taking off from cyborg anthropology) will be important in this. Some applications are emerging that are trying to do this more.
The approach to this is to understand the decisions and actions more intentionally so that they can be crafted better. E.g. the example from Target where they could predict that the customer is pregnant very accurately (model worked!), but you should not always act on it directly. You need to think through the empathic scenarios to decide when to act on that prediction vs. not.
It is very, very important to get educated on AI and software in general. A great book is “How to Speak Machine” by John Maeda.
AI business books (mine comes out from Wiley in Q2 of 2020). MIT’s AI Strategy online course. Andrew Ng’s AI course called “AI for Everyone.”
On the one hand, they are all horrible misrepresentations perpetuating myths. On the other hand, many are entertaining!
Deep Learning by Bengio, Goodfellow and Courville.
Bespoke AI models using Adobe DS Workbench and Sensei.
Communications Leaders Rally Around Artificial Intelligence, According to Global Survey
It is very, very important to get educated on AI and software in general. A great book is “How to Speak Machine” by John Maeda.