What Artificial Intelligence and Robotics Mean for Entrepreneurs
Entrepreneurs all know that Artificial Intelligence and Machine Learning are hot topics of focus among startups. But terms like ‘deep learning’ and ‘neural networks’ have been quickly absorbed into even the general public’s lexicon, and today it can be difficult to tell who really knows the technology they’re talking about, and who is simply pumping buzzwords.
Investors do have a genuine and growing interest in funding Artificial Intelligence ventures, but slapping these terms onto a pitch deck no longer does the trick for opening the funding faucets.
Recently, Founder Institute CEO Adeo Ressi sat down to discuss the future of AI and Robotics Startups and Venture Capital with a panel of renowned machine learning experts from Philips Health, Uber ATC, Carnegie Mellon University, and the Robotics Hub.
The discussion touched on many of the types of questions we hear from entrepreneurs across the globe who are curious about these emerging technologies. Below we have highlighted some of the key points from the conversation.
Identifying Signals from the Noise: A.I. Investment Dollars are Discerning
The most important action for aspiring AI and robotics entrepreneurs to take is to build up the domain expertise to truly and thoroughly understand the applicable technology.
Even for non-technical founders, buzzwords are no longer passable credentials: in order to be able to successfully articulate value to investors, entrepreneurs must possess real fluency in their professed technology’s domain. When an investor sees artificial intelligence or machine learning referenced in a pitch deck, the first question they will always ask is,
What kind of algorithm is your solution actually using here?
An answer containing a phrase like “supervised learning” would pass the investor’s initial test, because it implies the founder has a real conceptual framework around what their technology does - in this case, using a training data set to teach an algorithm to infer associations between an input and output. Replying instead with an answer like, “Well, we want to use machine learning to do XYZ,” would fail the investor’s initial test, because the entrepreneur is implying that they don’t yet have any real domain expertise.
Buzzwords do the trick for grabbing attention, but not even all software algorithms qualify as AI or machine learning. Investors obviously understand this, and are increasingly sophisticated in how they discriminate between different types of algorithmically-enabled technologies. According to featured panelist Jeff Schneider (Engineering Lead at Uber ATC & Professor at Carnegie Mellon), entrepreneurs should be driven by results and not simply driven to employ any particular technology,
Don’t get enamored of the method of AI. Instead, ask yourself the really hard question: Is there actually a product here that is better than what’s already out there?
Walking dinosaurs? Not quite, but many industries are bracing for impact.
While prominent voices have worried aloud about the future for Homo sapiens in a machine-dominated world, there is no immediate danger (yet) from any so-called ‘general AIs.’ Instead, industries and occupations poised for immediate disruption by artificial intelligence are those with the narrowest and most repeatable tasks - tasks that machines can easily be trained to do.
The expert panel identified a number of industries where narrow artificial intelligences are already delivering promising results, including healthcare, transportation, security, and finance:
Diagnostics - radiology and image recognition, understanding and predicting disease emergence and progression, and predictive analytics in therapy responses
Surgical robotics - assisting physicians, as well as nanorobotics
Personalized medicine - individualized pharmaceuticals, medical devices, combining patient history with genetic information and applying machine learning to predict individual patient outcomes
Drug discovery - pharmaceuticals, as well as sensors monitoring behavior in animal testing
Natural Language Processing - narrow and repetitive conversations only, like ordering a delivery or scheduling an appointment
Autonomous vehicles - trucking, ride hailing
Last mile delivery - drones, two legged walker
Vertical farms using AI to monitor and robotics to harvest in precision indoor agricultural production
Internet companies & User interfaces
Page layout optimization
Aesthetics ranking using computer vision
However, every one of the machine learning experts on the panel remains optimistic about a future that still includes plenty of work for human beings. According to panelist Michael Harries (Partner at The Robotics Hub, Associate at Creative Destruction Lab),
That human interface, there’s no reason to get rid of that anytime soon.
The Big Opportunities Are in the Data
In addition to Python and PyTorch, machine learning work is increasingly being pushed into modeling languages and probabilistic language frameworks, and programmers are even finding uses for older languages like Prolog and Lisp.
Entrepreneurs need access to data sets in order to train machine learning algorithms. But access to this type of training data can often present a barrier to entry for entrepreneurs, especially when accounting for other issues, like GDPR privacy, permission, and anonymization. It is possible to obtain such training data, and a first place to look should always be publicly available data sets. These data can be great for startup teams just getting started; however, they are also limited in scope. According to Schneider,
For the car companies, there are public data sets for that application; but all the car companies go out themselves and collect many orders of magnitude more than they could get any other way.
So how can entrepreneurs hope to compete, when the big companies have the resources to spend and invest in gathering data to train their own algorithms? According to panelist Birpal Sachdev (Leader of Advanced Technologies at Philips Health), entrepreneurs have the opportunity to create value complementary to the big tech companies,
Even if you’re a big company [like Philips], our disadvantage is that we are slow, in doing anything. And startups have the advantage that they can go fast. So collaborations among startups and big companies, I think that is the key. We will get you the data, you help us solve the problem faster.
Emerging Challenges Present New Opportunities
For emerging technologies like AI that are still only beginning to take shape, there is a wealth of new challenges and opportunities that arise from each advance made early on. As technologists learn to develop more advanced AI systems, there will always be a need for ever more advanced training, controls, and safeguards.
The expert panelists identified safety, validation, and verification of AI systems as an important new series of challenge that presents enormous opportunities. This is because AI systems are currently transitioning into the real world, controlling technologies that have ever-more consequential bearing on human lives, like those of autonomous vehicles. The risks of AI making a mistake is inconsequential if the technology in question is a chat bot, versus the potential for downside risk in AI piloting a driverless vehicle. This safety-critical work is just in its infancy, but will only become more and more important as artificially intelligent systems increasingly enter into the physical world of human life.
There will even be opportunities in so-called “policy innovation,” as governments whose knowledge always lags behind that of technologists seek to wrangle with the important task of shaping the regulatory landscape for these emerging industries. Entrepreneurs with established AI domain expertise will be critical in helping lawmakers to understand where to apply regulations versus where to let industry self-regulate as it matures.
Ethical questions too are beginning to be raised, and important work will need to be done to help ensure that AIs become agents for good, and don’t fall to the dark side. The expert panelists agreed that machine learning itself is neither inherently good nor bad, but that the machines are amoral: AI will only be catalysts to help humans do whatever it is that humans do, faster. For this reason, it will be crucial to make sure that AI does not inherit the worst of our human qualities, to ensure for example that training data sets do not teach AIs to learn the same prejudices that hinder humans from attaining our greatest collective potential.
Ultimately, the breakneck progress of artificial intelligence and robotics are already well underway. For better or for worse, the acceleration of these technologies is now inevitable. It will be up to this next generation of entrepreneurs to make sure that artificial intelligence technologies mature with human beings in the machines’ best interests. For now at least, human innovators are the ones in the driver’s seat, and it is up to us shape the ‘who’ that AI becomes.