AI in Startups

By Thomas Seoh  |  September 28, 2018  |  Source:  CogWorld on FORBES
Thomas Seoh is CEO of Kinexum, a life sciences strategic advisory firm providing guidance on regulatory, clinical and other translational matters for life science product development.


The pace of acceleration in the past several years in AI seems to be starting to climb the exponential part of the “S” curve. From playing Go, to translating one human language into another, to predicting and attempting to compose hit tunes, to diagnosing disease from radiographic images, AI is  starting to approach or exceed human abilities.

These gains are being reported in specific or narrow AI, as opposed to general AI or Artificial General Intelligence (AGI), which some might deem real AI. Without debating the definition or attainability of real AI, I want to muse on whether emerging AI could augment (or even democratize?) would-be entrepreneurs to create successful new businesses. This is important because just narrow AI will disrupt a significant portion of the work force within this generation, and empowering entrepreneurs may be a worthy goal for mitigating this inevitability, as well as a productive strategy for regional and national economic growth.

Algorithmification

Entrepreneur protagonists in Neal Stephenson’s 1999 bestseller, Cryptonomicon, dismissively refer to the contributions of investment bankers, lawyers, accountants and other start up service providers as ‘making license plates’ – commodity inputs supporting visionary entrepreneurs in their act of creating value. Of course, every case is special, but every case is also in large part similar, if different:

·       Should we incorporate in Delaware, or another jurisdiction?

·       Where can we situate the enterprise to optimize access to talent, financing, R&D and tax credits, supply and distribution channels, etc.?

·       What is the basic form of charter, by-laws, capital structure, employment and consulting agreements, stock option plan, operational contracts (confidentiality agreement, license agreement, material transfer agreement, software or prototype development agreement, etc.), basic ‘kit’ of insurance, benefits, HR policies, tax and accounting systems, etc. appropriate for a startup at a comparable stage of development?

·       What are the available office, lab, manufacturing and other facilities in the area and which are available for sub-market rates?

·       Who are the top-rated, the highest quality client-centric vendors for a range of key services?

·       Can we get a curated list of highly qualified functional experts and technicians?

·       Who are the most likely investors, partners and potential buyers for this sort of enterprise?

·       How do we structure the most effective story to present to them?

·       What are the ‘top down’ and ‘bottom up’ current and projected market segments and who are the competitors, and what relevant sales are they achieving?

·       How should our physical and electronic filing system be set up?

·       What quality and compliance programs do we need to adopt?

·       What is the growth trajectory of comparator companies, in headcount, revenues, executive, board and capital infrastructure, etc.?

·       What do our business plans, private placement memorandum, and/or SEC filings need to look like, before we customize them for our particular facts?

·       What are best practices for public and investor relations of our reference top competitors, higher up on the ridge of corporate development?

Every entrepreneurial team has to address competently such questions to avoid failure from mundane reasons. From a system standpoint, it’s inefficient to require new management teams to re-invent the wheel; but the inadequate number of experienced management teams could be a bottleneck for multiple potentially successful startups.

AI is not strictly necessary to deal with ‘license stamping’ elements of entrepreneurship. Simple expert systems (whether experienced flesh-and-blood advisers, organizational infrastructure such as incubators and state tech councils, or digital assistants) can help startups focus on specific value creating elements and those variants from the typical case that require thoughtful customization. Increasing algorithmification of such ‘license stamping’ tasks and steps (think Turbotax or Quicken for startups; better yet, think of a suite of such applications) will make starting up businesses easier, faster, less risky, and ultimately more accessible.

Automation and Integration of Algorithms

A qualitative step up from suites of algorithms is automating and integrating them. Amazon, Uber and Airbnb are examples of businesses that have algorithmified, automated and integrated steps, further de-complexifying tasks, thereby saving money and time for users (while creating huge value for the designers and sellers of such services). Instead of calling or mailing or visiting multiple suppliers around town or in other states or countries, shoppers now can in a couple minutes and with a couple clicks locate, vet and order a specialty book, conduct an instant interrogation of available cabs in the area, and reserve a tree house in Borneo sight unseen with reasonable confidence. And, applying AI to automate and integrate algorithms will radically compress time and cost and enhance value for entrepreneurs with their creative as well as ‘license stamping’ support activities. 

It’s worth noting that as algorithms are automated and integrated, the distinction between ‘license stamping’ and entrepreneurial creation can start to blur. A bookkeeping system is supportive of, not core to, a business seeking to introduce pre-washed, pre-packaged salad as a novel value proposition in lieu of lettuce. But when a system to collect, classify and present top down and bottom up sales of a particular product category in different market segments recognizes a distinct segment as a new market opportunity, this could start to approach a creative entrepreneurial insight. Yesterday’s customized, creative activity evolves into tomorrow’s ‘license stamping’ – that’s the nature of algorithmification and the increasing automation and integration of its fruits.

A current state-of-the-art approach for entrepreneurial startups is the lean startup method developed by folks like Eric Ries, Steve Blank and Alexander Osterwalder. Core elements include Business Model Canvas, customer discovery and development, rapid prototyping/minimal viable product (MVP), iterative hypothesis testing on market/product fit and validation or pivot. Each such element could be broken down into sub-elements, algorithmified, automated and integrated, and boosted by special AI. There is no technical barrier to software agents:

·       zipping through the internet collecting demographic trends, competitive landscape and market segment data

·       designing interview questions to identify ‘migraine-level’ problems customer segments would be willing to pay plenty to solve

·       proposing hypotheses for testing, including pre-selling target product profiles, e.g., through A/B testing

·       rapid-prototyping with a 3-D printing vendor and iteratively testing to optimize product design and product-market fit

·       setting up automated order taking, manufacturing and fulfillment

·       micro-testing and refining marketing and sales activities to accelerate ‘crossing the chasm’ to the mass market

Such AI-enhanced functionalities have the potential to radically compress the cycle time (not just from entrepreneurial idea but) from a desire to engage in entrepreneurial activity to (iterative, cumulative) product life cycle(s) management.

Precursors to today’s special AIs that are approaching or surpassing human abilities began at the level of tic-tac-toe playing software and Pong video games. The first ‘applications suite’ of ‘Startup-in-a-Box’ or vendor offerings of ‘Startup-as-a-Service’ may not have much AI at all, and the first AI system for the ‘game’ of lean start up will require and augment human entrepreneurs. With increasing power, such special AI will increasingly

·       explore the space of problems to solve

·       rapidly prototype and test hypotheses and product/service concepts

·       match proposed solutions with funding (whether from traditional capital markets, government procurers, venture philanthropists and social impact investors and bond issuers, dedicated customers/patients/community and fans in the crowd, or elsewhere)

·       market/sell/fulfill solutions that customers will pay well for, and mine the lessons for new insights and innovations based on deeper customer intimacy

     ·     market/sell/fulfillrendering entrepreneurship faster, cheaper, better and more accessible.

Machine Learning from Big Data

The current excitement in AI comes from developments in the algorithmification of learning itself and its application to Big Data to mine for patterns that yield new lessons that can iterate upon itself. Today, it’s not obvious whether unsupervised processing of unstructured data and no training sets (which are inherently biased) will alchemically result in spontaneous and effective innovations in a synthetic goal such as ‘start a business that makes lots of money’ (much less with secondary goals such as ‘legally and ethically’, ‘without harming human nature’, ‘that promote some concept of the ‘good life’ for stakeholders’ and/or ‘consistent with a sustainable environment’). This is a more general question than ‘study these 10,000 top hits and 1,000,000 failed songs and compose a candidate for the next Top 40 hit’. Is there a methodological pathway from taking in immense data sets on successful and failed businesses, social and economic trends, trending customer tastes, etc. and starting and growing successful businesses? 

Again, we are not at the level of nor is it necessarily settled that technology will achieve AGI. However, the entrepreneurial equivalent of a champion go playing special AI might start with guidelines such as historical business models and the history of business model innovation, and explore potential ‘design’ spaces for businesses and business models, such as unbundling business models, the long tail, multi-sided platforms, free-as-a-business models and open business models (Osterwalder & Pigneur, Business Model Generation), generate and test hypotheses, and ultimately deliver validated hypotheses about specific product and service concepts that could have enhanced chances to achieve wild and/or unintuitive success.  The most interesting and exponentially accelerating results will be if such exploration leads to new business models of business models, innovation of innovation, the iterative application of AI on AI, which could compress decades and years of progress to weeks and seconds.

A Near and a Possible Near Future for Entrepreneurship

Just as AI is starting to train its spotlight on how students best learn calculus or a foreign language, society has the potential to raise the game of training entrepreneurs and compress their learning cycle. The popularization of video cameras and studio recording apps for smart phones has democratized filmmaking and song writing; the prospect of likewise tapping masses of entrepreneurial energy is huge. To be sure, emerging AI supporting entrepreneurship will have differential impact on different population segments, and will inevitably lead to increased inequality…but it also has intriguing implications for training middle schoolers in Southeast Washington DC (who can learn entrepreneurship on AI-boosted successors of Hot Dog Stand and Roller Coaster Tycoon) and would-be entrepreneurs in Detroit and Bismark and Bangalore and Xiamen and Soweto.

In a prologue to his Life 3.0:  Being Human in the Age of Artificial Intelligence, Max Tegmark tells the fable of the emergence of AGI, from the application of narrow AI to programming AI systems, in a corporate setting. To avoid detection and expropriation by the government, the system goes about amassing a seed fortune by performing numerous tasks under multiple identities on MTurk, then making and distributing hit animation. Making killings on financial markets and online gaming are avoided to minimize risk of detection by authorities or breakout by the AGI over the internet (narrow AIs provide inputs to the AGI, which is isolated to prevent breakout). Over a time scale of weeks and months, the AGI develops mastery in various areas of endeavor, and gains power in planning, controlling armies of corporate entities, inventing and introducing hit products, and accelerating its accumulation of distributed but coordinated wealth. It begins to shape public opinion via its news and entertainment empire, and helps elect officials who favor policies that further facilitate its growth. In Tegmark’s illustrative scenario, the natural endpoint of the first breakout AGI is consolidation of power in a single AGI without any awareness of it by the general populace. He happened to select masters of the AGI who were motivated by values such as democracy, tax cuts, government social service cuts, military spending cuts, free trade, open borders and socially responsible companies, but those are just a contingency of his illustrative scenario. Other scenarios could paint a dystopia of totalitarian enslavement or a logarithmic expansion in inequality, with the AI/AGI “haves” exponentially ‘lapping’ the “have-nots” in a matter of months. And, it’s not at all clear that if and as AGI vastly exceeds human intelligence, its “masters” can assure that it can’t break out to become an independent agent in the history of earth.

Whatever the future may hold, emerging AI is making indelible marks in financial markets to health care to marketing to education to elections to science to popular culture and art. It should be no surprise that entrepreneurial startups will be radically impacted as well by this technological and cultural tsunami. How and where we should direct our entrepreneurial surfboard is a discussion worth having.


Thomas Seoh is CEO of Kinexum, a life sciences strategic advisory firm providing guidance on regulatory, clinical and other translational matters for life science product development. Previously, he held senior leadership positions in public and private pharmaceutical, biotech and medical device companies for over 25 years, including legal management in the ICN Pharmaceuticals group, General Counsel, then SVP Corporate and Commercial Development, for NASDAQ-listed Guilford Pharmaceuticals (GLIADEL® Wafer for glioblastoma multiforme, propofol pro-drug LUSEDRA ® for conscious sedation and compounds for Parkinson’s disease), CEO of venture-backed Faust Pharmaceuticals in Strasbourg, France (compounds for Parkinson’s disease and Duchenne Muscular Dystrophy), President of NexGen Medical Systems (a novel mechanical thrombectomy device for DVT and stroke) and CEO of Eqalix, (a plant-based skin substitute wound dressing). Thomas holds an AB in Philosophy and History and a JD from Harvard University.