Notes on Engineering Health, October 2024: Notes on Risk, Uncertainty, AI & Entrepreneurs

As venture capital investors, we like to believe that we are in the business of managing risk. We evaluate potential investments by doing due diligence on a range of risks including the technical risk of whether a proposed technology can in fact be engineered, the market risk of whether a new technology will take market share in a given category or create an entirely new market to be exploited, and the management risk that this group of entrepreneurs can solve both the technical and market problems in front of them. In each case, we generally act as if these risks can be assessed probabilistically—that is, on a set of calculable odds. In particular, we most often act as if we are dealing with Bayesian risk, relying on knowable and known probabilities that we constantly update as our due diligence work is bolstered by new information.

But, in the world of entrepreneurship, much of the game is played in an environment of uncertainty rather than risk. True uncertainty involves situations that are non-computable because we lack the information to calculate odds accurately. This is the land of the unknown and unknowable in which probabilities cannot be meaningfully assigned and thus the idea of probability itself becomes meaningless.

The distinction between risk and uncertainty in the business world was first delineated by University of Chicago economist Frank Knight in his 1921 book Risk, Uncertainty, and Profit (cited below as RUP). “There is a fundamental distinction between the reward for taking a known risk and that for assuming a risk whose value itself is not known,” Knight wrote. A known risk is “easily converted into an effective certainty,” while “true uncertainty is not susceptible to measurement.”

Knight breaks down uncertainty into a meshed set of knowledge and action problems that arise when an entrepreneur does not know whether something will be possible in the future, and therefore cannot know or predict accurately the future state of a given decision environment. For example, uncertainty affects many crucial entrepreneurial decisions, from choosing which products to develop and markets to pursue, to designing a business model that will meet future goals.

In particular, Knight identifies four correlative problems (which are summarized in detail in this recent paper by Townsend et al):

1) Actor ignorance
“The world is made up of objects which are practically infinite in variety...and when we consider the number of objects which function in any particular situation, and their possible variety, it is evident that only an infinite intelligence could grasp all of the possible combinations (RUP, p.207).” This is a knowledge problem that occurs at the actor level as an entrepreneur can never know for sure what possibilities may exist (or not) in a future decision environment.

2) Practical indeterminism
Another knowledge problem that encompasses the fact that in the real world decision-makers constantly face multiple possible future outcomes that cannot be known in advance: “The postulates of knowledge generally involve the conclusion that it is really determined in the nature of things which house will burn, which man die, and which face of the thrown die will come uppermost. The logic which we actually use, however, assumes that the result is really indeterminate, that the unknowable causes actually follow a law of indifference (RUP, p.219).”

3) Agentic novelty
The actions entrepreneurs take to introduce new possibilities into a system that diverge from existing approaches, which actions in turn actually cause the practical indeterminism described above. This is an action problem tied to the actions (and non-actions) undertaken by an entrepreneur.

4) Competitive recursion
Another action problem which involves that consequences of the novel choices and actions of entrepreneurs to intervene or not, and how the consequences continuously and recursively feed back into the decision environment. This problem encompasses the difficulty of twofold inference: entrepreneurs “must infer what the the future situation would have been without our interference, and what change will be wrought in it by our action (RUP, p. 202).”

Interestingly, the problem set described by Knight provides a strong view on computational limits of the current dominant approaches to Artificial Intelligence. As Townsend et al write:

For Knight (1921), the knowledge problems entrepreneurs face in ‘knowing something about the future’ are rooted in the indeterminism of the environment – the fact that future states of the world can evolve in dynamic and unpredictable ways. In this sense, the ignorance of the entrepreneur about the future and the limitations of probability theory are not temporary states that can be resolved through inductive processes of knowledge acquisition or through the development of better probabilistic tools.

To the extent that modern machine learning techniques are driven by probability calculations, Townsend et al conclude:

“[N]o matter how much information the entrepreneur collects or generates, computing whether a future state is possible or impossible will only ever be, at best, partially predictable due to the dynamic nature of decision environments. … [T]hese technologies will always be limited because the underlying generative process ultimately relies on a stochastic function to generate new possibilities.”

A world of unknowability by definition requires different strategies to a world dominated by knowable risk calculations. And, thus, for entrepreneurs, scenario thinking, portfolio approaches, flexibility and adaptability, resilience, judgement and preparedness (see commentary here) remain crucial in seeking success.

And it is somewhat comforting that it appears the role of the entrepreneur will never be subsumed by (at least the current generation of) machines as the process of pursuing new and novel businesses “fuels a never-ending process of generating novel solutions to emerging problems that virtually guarantees that the futures which emerge will often deviate in important and unknowable ways from the present (Townsend et al).”

– Geoffrey W. Smith



First Five
First Five is our curated list of articles, studies, and publications for the month.

1/ Biopiracy
An international battle over who owns the natural world’s genetic data—and who should benefit from the multibillion-dollar discoveries derived from it. Read more here >

2/ The fourth dimension
What makes 4D weirder than all other dimensions? As one mathematician [said], “In dimension 4, everything goes a bit crazy.” That’s because, according to another [mathematician], “there’s just enough room to have interesting phenomena, but not so much room that they fall apart.”
Read more here >

3/ Information theory and evolution
How systems self-emerge and self-configure for information exchange from 0 to 1 to n bits.
How these systems necessarily culminate in the complexity and diversity of living things as a result of rules governing information theory, where natural selection is a specific case of the laws governing noisy information exchange between finite sized systems. Maybe? Read more here >

4/ Organizing for innovation
An argument for innovative organizational design based on Bolt, Beranek & Newman (the original ARPAnet contractor) and some problems this model is optimized to pursue. Read more here>

5/ Making sense of chaos: a better economics for a better world
“We live in an age of increasing complexity, where accelerating technology and global interconnection hold more promise – and more peril – than any other time in human history.

The fossil fuels that have powered global wealth creation now threaten to destroy the world they helped build. Automation and digitisation promise prosperity for some, unemployment for others. Financial crises fuel growing inequality, polarisation and the retreat of democracy. At heart, all these problems are rooted in the economy, yet the guidance provided by economic models has often failed.

Using big data and ever more powerful computers, we are now able for the first time to apply complex systems science to economic activity, building realistic models of the global economy. The resulting simulations and the emergent behaviour we observe form the cornerstone of the science of complexity economics, allowing us to test ideas and make significantly better economic predictions – to better address the hard problems facing the world.

In this talk economist Doyne Farmer presents a manifesto for how to do economics better.” Watch here >



Did You Know?
In this section of our newsletter, we seek to demystify common terms and practices in our work as investors.

Cap Table
A capitalization table, or cap table, is used to show the equity capitalization for a company and helps a private company main a calculation of its market value.  While all companies can use a cap table, they are particularly necessary for startups and early-stage companies that have or will receive venture funding.

Cap tables detail who has ownership in a company and list all the securities or number of shares of the company such as common shares, preferred equity shares, stock warrants and convertible equity.  Overall, a cap table shows the total market value of a company and its components. As a key point of reference for entrepreneurs and investors, the cap table is considered in every financial decision that has an impact on market capitalization and the company’s market value. As such, cap tables must be accurate, and kept up-to-date based on the most current information.

– Haiming Chen & Dylan Henderson

‍

‍