IAI. innovative augmented intelligence 🤖

[hyunjimoon]

• innovative:= making new thing beneficial to others
• most fundamental architecture as ai-centered perspective can be helpful for human-ai collaboration. this can be directly applied to Bayesian Entrepreneurship: 🗺️map for startup compass🧭 #159 where individual"s" collaborate with ai (ai as startup mentor), FOW: future of work T4 in M3S, ai for human capital and economic growth #161 where individual collaborate with ai on developing itself (education, career path)
• theory: ai's bias mitigation strategy and how human can complement it
• tool: PPL, gen-finance

[hyunjimoon]

why we need ai: too complicated! (comment from scott's entpre.strat. lecture)

[hyunjimoon]

I have questions on largely two topics: o4s (operations for startup) and o4e (operations for entrepreneur). framing startup as group of entrepreneur"s", we first characterize startup's growing process and list conditions for individual (both capability and agility to act upon different signs in an aligned way with the rest), we can understand how one "becomes" an entrepreneur.

I'm approaching o4s with unsupervised learning to cluster phase (1.1) and industry group (1.2) and o4e with supervised learning on exposure to startups via NUS enterprise (2.1)

1.o4s

1.1 phase

Q1. how to define/measure the phases of startup growth? could there exist some time series characteristic (e.g. revenue exceed cost) that is generalizable enough across different industries? Example of phase:

• Charlie Fine (advisor)'s nail, scale, sail here
• NUS entrep.'s: early stage hanger, growth stage (block71, nus sing.science park, agricenter) from its website banner

Angie attempt:

• make analogy between learning phase in bayesian workflow (iai1), probabilistic programming (iai2), operations for startups (o4s), entrepreneurs (o4e) individuals concrete, to apply fast clockspeed theory developed from gen-finance #177, IAI. innovative augmented intelligence 🤖 #174 to slow clockspeed management in Bayesian Entrepreneurship: 🗺️map for startup compass🧭 #159, FOW: future of work T4 in M3S, ai for human capital and economic growth #161

1.1,2 iai1, iai2

• below fig1.1 is iai2 which has resemblance with o4s's nail it (=unstructured learning) is structure learning (part 1) and scale+sail it is exact inference (part 2). o4s's scale it focus on feasibility (iai: accuracy) , sail it focus on viability (iai: resource efficient ai which is in line with T3 in M3S (which angie is involved with))

• fig1.2 nail: $P_\mathcal{L}(s,\theta)$, scale: $P_\mathcal{L}(y|s,\theta)$

1.3 o4s

• startup is black box and is surrounded by different external evaluators like mentor, customer, investor,

1.4 o4e

• failure function where bottleneck change among three components: early learning (burn in), random failure, wear out. likewise, phase from scale to sail can be defined by revenue first exceeding cost.

• o4s and o4e's quality perception of a startup after several customer interviews, quality perception of oneself after several job interviews

1.2 industry

For scalable computational representation of entrepreneurial decision making, I'm generalizing this demo) which simulates revenue and cost time series of adtech startup then provide its inference.

Q2. how can i cluster revenue and cost time series using prior knowledge on industry group? e.g. for [phase2 x o4s] (phase2 as revenue most likely occurs from phase2)

Angie's attempt:

• cluster revenue and cost time series of different tech companies: AgTech, App Tech, AI Tech, BioTech, E-Commerce Platforms, Retail Technology, EdTech, CleanTech, FinTech, Tech Hardware, Health Tech, IT Tech, Cloud Computing, Digital Currency, Digital Media, Telecommunications, Mobile Tech, Audio Tech, Cyber Security, Legal Tech, PropTech, Digital Marketing, Engineering Tech, Software Tech, AutoTech, Travel Tech, Video Tech (based on BAE. bit atom energy management 💡⚙️⚡️ #147)
• bringing in methodologies from Feras' (advised by Vikash) thesis pdf may be helpful. Specifically, apply fig1.5 which structure-clusters economic indicator (GDP per capita) time series (50 years). example of structures are linear trends, exponential growth, and changepoints.

2.o4e

2.1 NUS ENTERPRISE

Q3. would getting data for [phase1 x o4s, o4e] be possible to test hypothesis of "more exposure to innovation, the more entrepreneurship one person gets"? we can compare entrepreneurs who've been abroad VS who haven't been abroad.

[hyunjimoon]

• phase123: structure then parameter fitting learning
• using phase123, pivot from high to slow clockspeed industry to learn faster: need to transfer learning and detect phase1 termination diagnostics
• e.g.1 political advertisement to luxury brand, e.g.2 digital to physical (?)

ad tech: luxury brands are slow clockspeed but higher lifetime cost - learn business configurations (not value but structure, nail) in high clockspeed industry like political advertisement, lower supply side uc then pivot to slow clockspeed

[hyunjimoon]

@miaomiaozhang20, Angie first meeting
with the purpose of pushing 15357_proposal.pdf forward.

1. ⚙️ building ai agent for lit.rev
2. 📚 ai readings - discuss documentation

m2 introduce: Agrawal, A., Gans, J., & Goldfarb, A. (2020). The Economics of Artificial Intelligence: An Agenda. The University of Chicago Press., Arora, A., Fosfuri, A., & Gambardella, A. (2004). Markets for Technology: The Economics of Innovation and Corporate Strategy. MIT Press.
Angie introduce: Agrawal, A., Gans, J., & Goldfarb, A. (2022). Power and prediction: The disruptive economics of artificial intelligence. Harvard Business Press.

3. 💬 prize vs patent; locus of uncertainty as in FOW: future of work T4 in M3S, ai for human capital and economic growth #161 (comment)
   is crowd sourcing more close to prize or patent? I think prize, as need is specified.

prize (system specifies a specific need and N number of people find solutions): if a policy maker knows what need is (at least to the level of forming a efficient question), larger N is better given heterogeneity among N is big enough, such that benefit of one additional agent outweighs the cost for the loss of overlapping work. we'd want to increase diversity of entering capability to prevent dual work between two agents. e.g. artist's vs scientist's process to solve given vaccine problem may differ. We can learn strategies to design better index using $N_{eff}$ effective sample size where $\rho$, thinning, FFT would all have analouge.

$$\hat{N}_{\text {eff }}=\frac{M \cdot N}{\hat{\tau}}$$

$$\hat{\tau}=1+2 \sum_{t=1}^{2 m+1} \hat{\rho}_t$$

patent (system tests need-solution pair): if a policy maker has high need uncertainty, testing need-solution pair might be more efficient than trying to limit the need that may narrow the candidate solution space (which inevitably resembles affordance space of given need).