Technology

Machine thinking is here

You don't build an AI executive by mindlessly parroting stolen ideas from the internet like other generative AI models. You need to steal heaps of ideas and then pick the best one. In short, you need machine thinking. So we invented it.

Three-pillars of Ceobot's machine thinking model

1. StrategiParrot idea aggregator

2. GreediRat decision-maker

3. MilkMonie arrogance management system

StrategiParrot idea aggregator

We’ve used this basic model for Ceobot. Instead of using the internet, we trained our model on 75,000 annual reports, which detailed the strategies of thousands of publicly traded companies. Working from this massive dataset, the machine can regurgitate a validated profit-maximizing strategy for just about any scenario. This is StrategiParrot.

You've heard of the stochastic parrot model of generative AI. It takes a prompt like “Why does baking soda react with water?” and reorganizes those words into the root of a statement, “Baking soda reacts with water because…” Then it finds all articles with those words and chooses the next most common word in the sequence; continuing on until the sentence, paragraph, or essay is complete.

GreediRat profitable pathways decision maker

Some of the most ubiquitous experiments in psychology are rat mazes, where rats are incentivized to find their way to the end in exchange for a reward. It’s usually rat food as that’s what they seem to like.

Thousands upon thousands of these studies have confirmed that rats will repeat actions which increase personal rewards. They'll choose a path with two treats over a path with one.

This is rat learning. Machines can do this.

We trained an algorithm to run a rat maze, assess the various rewards on offer, and then identify the path with the most to offer. After running the simulation thousands of times, we’ve achieved a model with near-perfect rat status. It identifies the most profitable pathway 99% of the time.

This represents a move from cut-and-paste generative AI that mindlessly regurgitates other ideas from the internet, to thinking AI that assesses multiple options and chooses the best.

We modified the original version of GreediRat to pursue piles of money rather than piles of desiccated entrails. Bingo, bango, we had a profit-bot.

StrategiParrot feeds in stolen ideas, and GreediRat chooses the one that offers the most spoils.

Don’t be fooled by the modesty of the animal on which this system is modeled. If you don't think that rat-level thinking is impressive, then you don't think corporations are impressive.

MilkMonie arrogance management system

One of the core challenges of this technology is hubris. Given its training material, Ceobot immediately assumes it's in charge of an S&P 500 company that can treat customers, employees, suppliers and regulators with disdain.

That gross sense of entitlement isn’t a bad thing, per se. Being a dominant monopolist is definitely one of the best ways to maximize profits. But you need to get there first. Not all of our software users will be market giants who can do whatever they please.

We had to build an arrogance thermostat that would stop Ceobot from veering toward total corporate anarchy (until it had the social license to do so).

So how do you stop your software from being a dickhead (all of the time)?

Programming a conscience is tricky and requires moral choices, which is fraught. Not to betray our Silicon Valley nihilism, but morality is a human illusion and we didn’t want to constrain our technology with subjective values.

The world created neo-liberal economics specifically to avoid morality. Letting the market decide what’s right or wrong is far more objective than getting into the messy business of culturally-mediated ethics. We simply don't have the right to impose so-called 'principles' on laissez-faire economics.

Instead, we needed a market-based gauge that helped the software know when it was wise to be subservient to other stakeholders – such as suppliers, employees, customers and governments – versus when it’s free to ‘enable all its learning’. With that in mind, we built the MilkMonie arrogance management system.

Finding a model on which to build MilkMonie

Our first instinct was to treat the market as a commercial democracy, where customers vote with their wallets and popular companies ascend to power.

We duly programmed Ceobot to switch to full profiteering mode only after the company under its leadership became a market leader. But after running simulations with this model, we found the software was slower to start gouging customers than real-world CEOs.

We were clearly missing some nuance in how companies assumed a mandate to exploit their market. The search for a new paradigm was long and extensive. We ultimately settled on a less celebrated but arguably more ubiquitous sociopolitical arrangement than democracy.

Every school lunchtime, bullies of the world exert control over a reluctant cohort even though they may not be liked by a single member of said cohort. And once ascendant, the bully extracts rent (aka, milk money) simply for not making things worse.

This seemed to be an accurate analogy for the relationship between corporations and their stakeholders. In particular, the bully’s clarion call – What’cha gonna do about it? – aptly captured the dilemma of a customer (or employee or supplier) who’s been enslaved by a big corporation.

Customers, in particular, become trapped because the switching costs are too high. They simply can’t bear the burden of phone trees, break fees, and loss of data that characterize the torturous exit from an incumbent supplier – only to go through a similarly exhausting onboarding process with a new company.

And then it hit us: the new provider is the problem, too.

A bully, you see, is not an individual. It’s a fixture. A phenomenon that’s as inevitable as playgrounds themselves. Escape from one bully invariably leads to capture by another. Even kids know that. We just needed some math to model the intrinsic hopelessness of the market. Enter the TCF.

MilkMonie's tacit collusion framework (TCF)

Customers won't break ties with an extractive corporation if it merely exposes them to capture by another.

The failing of our beta model was to assume consumers had genuine choice when, in fact, they don't. Everything sucks. Nietzsche said it best: "The free market is dead." Ceobot wasn't properly leveraging that fatality.

MilkMonie needed to acknowledge that decisions to exploit a market aren't driven by a single market leader, but are collectively made by whole industries at a time.

So we created a category tracker that monitors the deteriorating behavior of industry verticals.

It measures how extractive each industry sector has become by measuring things like price, privacy, responsiveness, service levels, and so on. All Ceobot has to do is make sure it doesn’t surpass the accepted rate of value decline for the industry it’s in.

We called this the Tacit Collusion Framework (TCF). You can, and should, be on the leading edge of value decline for your industry. But you can’t surpass it by more than one standard deviation. We’ve now built an index for measuring value decline in all major industries and baked the TCF into Ceobot’s MilkMonie arrogance management system.

The software is now perfectly arrogant.