N+1 Singer recently hosted a talk by Dr. Hermann Hauser FRS, KBE at the Walbrook Club on the current state of play within the artificial intelligence space. A serial entrepreneur, tech-focused investor and co-founder of Amadeus Capital Partners, Dr. Hauser is uniquely positioned to offer insights into the still embryonic AI sector.
The talk began with a discussion around where we are in terms of AI development and computing. Dr. Hauser started by drawing comparisons between the qualities of current transistors with the neurons in a human brain. Of concern for the luddites, transistors have already surpassed our neurons in terms of size (1,000x smaller), speed (1,000,000x faster), capacity (5x higher) and processing power (1,000x bigger). Furthermore, the advancement in audio and video available in robotics enables data collection on a much wider level. It also opens up potential for future AI to experience a much richer and more varied sensory experience than their human counterparts.
Dr. Hauser did note human beings continue to enjoy a lead over AI in the number of neural connections (100tr versus 1bn hosts/20bn IoT connections). Humans also, and perhaps more importantly, have an advantage in terms of energy intake; consuming 10,000x less energy than computing comparatives at just 20W.
Whilst the talk made clear AI’s capacity, successfully defining what quantifies machine learning, its potential uses still remain challenging. The Royal Society ML report classifies machine learning as ‘systems that learn from data rather than following pre-programmed’. Dr. Hauser further subdivides ML in to 4 separate branches:
• Supervised ML: Trained on labelled data (Robotic Process Automation (RPA), or business process automation).
• Unsupervised ML: Trained on data without labels (Complete functions on varied datasets).
• Reinforcement ML: Learning from experience, reward function (operations within human framed parameters, human oversight function required).
• Inverse reinforcement ML: System has to deduce reward function by observation (driving cars, data analytics and new data trend insights without human influence, cognitive learning).
We note Robotic Process Automation (RPA) or supervised ML is already seen in a number of ‘AI’ offerings available to consumers. RPA’s advantages to companies include a step-up in the capacity of staff, as well as the potential replacement of some low-level, easily automated roles. Yet Dr Hauser believes computing is now moving beyond binary ad deterministic computation (0 or 1), and into probability, statistically driven processing. This leads us to believe humans are likely to be needed less and less for oversight purposes as we progress beyond RPA.
So how might investors gain exposure to AI today? We see two potential investable pools:
First – companies which drive advancement in RPA, for example those which are currently introducing a heuristic learning element into the automated process. Human oversight is requested by the automated platform whenever a conclusion is unclear, meaning new parameters or processes can be added over time. Dr. Hauser refers to this educational element in his talks, describing the move from programming to learning. The move to learning empowers the everyday operators of a business to improve processes by dealing with platform queries themselves, rather than outsourcing to tech-focused developers who are likely to be less familiar with the business process. The widening of access to those without a tech background is likely to prove a material driver of value in our view.
The second pool includes companies developing hardware or software likely to drive AI advancement. We believe sensors and data gathering tools (improving quantum and quality of data; reduced security/fraud threat); data aggregation, translation and interpretation tools (improves manipulation and analytics); and software/hardware improving speed, flexibility and depth of processing capabilities (driving evaluation and creativity of AI) are areas of interest.