Fred Ehrsam, I would frame the problem differently. The need is not just for creating most powerful artificial intelligences, it is about redesigning the marketplace that can nurture such an ecosystem. The marketplace cannot be taken for granted. A poorly designed marketplace means poor AI, even if we assume that the state of AI today is structurally stable.
Incentivization, zero knowledge, data encryption are just a few aspects linked to efficient marketplace thinking. Web 3.0 needs to figure out Tim Berner Lee’s quest. For AI to be relevant it has to facilitate a Web 3.0 environment which means show utility beyond an isolated domain. This is where business models like Numerai are weak as they don’t facilitate or showcase generalization. Touting performance in an automated high-frequency world is old tech and low-frequency automation is a pre-step before any generalization that could aspire to be a master algorithm.
Hence before we even talk about incentivization or stake slashing we need to understand that model building will remain a broken science till we hard code the reference. A model needs a reference e.g. SPY is based on SPX. It’s an index comparison. If a US model has 100 blue chip stocks, it should be compared not to SPX but to the same 100 blue chip stocks in a zero IP context, zero IP being an MCAP weighted benchmark. The new marketplace needs the foundation stone of reference accountability before we address the AI challenge of generating more than 1% risk-weighted excess returns above the zero IP benchmark for near $ 100 trillion in investment assets today.
Even if the AlphaBots are eager to bet their reputation on verification, there is a need for education for the market players. What is being verified? How generalized is the model? How many domains does the model address? Is the methodology reusable? Is it an SSR (Systematic, Scientific, Replicable) process? Is this verification interpretable by an investor? Is there an offline reputation at stake? Is there a second layer of validation? Performance attribution is not science, it’s a set of rules that can be easily passed through blockchain to assist global alpha generators.
Backtesting, for example, can be addressed by testing the model for starting point bias, running the same methodology for different periods. How many of the models have seen taking the starting point bias test? If you want to invest your money on a week of trading performance, you should read the law of ruin. Performance attribution is not only for financial models, it is also for non-financial models because any nonfinancial data can be converted into a financial portfolio. This is why it’s easy to calculate the value of a particular data set. Hence a community of third-party validators and revalidators is essential before the verifiable computation overlay. The claim that third parties can introduce complexity and hence reduce ease of use is wrong because complexity is what’s stopping the blockchain from scaling.
A marketplace is not designed for power. It’s about a level playing field which allows for redistribution of power. It’s also about intelligence exchange and risk transfer. The intelligence has to be for everyone on the chain or off the chain. The intelligence should be measurable and dynamic. Intelligence which understands that intelligence extraction technology is more important than data.
A marketplace needs a vision as it was short-sighted rewards that led to the ongoing collapse of the crypto bubble. This marketplace also needs to think beyond tokenization. Otherwise, it’s like creating out-performance and losing it to currency risk. All tokenization is unintelligible if the same AI that generates alpha can’t think about a stable currency. We also need to have bigger aspirations when it comes to problems to solve and need to think beyond recommendation systems.
Preference and smartness are two islands. Preference always trumps smartness in the legacy world. Why do you think investor continue to suffer underperformance? Blockchain will remove some of this inefficiency as smartness will be more accountable, but preferences are not going away. Alpha objectivity is a lot different than the Google search functionality and the new marketplace will be about intelligence customizability and hence more pixelated preferences. Money has stickiness and it does not unstick that easy. Moreover, the search for alpha has a cost, and on average, dumber money (passive investing) has proved to be smarter than smart money (hedge funds). The blockchain marketplace can reduce this inefficiency but it won’t eliminate it.
An effective marketplace should not focus on dismantling data giants. Because decay emerges from the old generation thinking of the incumbent which was a startup struggling for capital a decade ago. Now the same giants are invested in their web 2.0 thinking and are too big to unlearn or unwind. Bracing for relentless change does not need just innovation, it needs escape velocity to survive. Hence, the only thing the new marketplace visionaries need to do is to embrace the change without any fear.