Botpower: a Universal Metric for AI
I have always found it interesting how certain measurements transcend the epochs.
The mile derives from a Roman soldier’s thousand paces - mille passum - as measured by his every other step. That is to say, the total distance of his left foot hitting the ground 1,000 times. Note this would be at a quick trot, not a leisurely walk.
A furlong (meaning furrow length) was the distance a team of oxen could plough without resting. This might seem quaint today except when you walk north-south on a city block which is… one furlong long.
And then there’s horsepower. I’d like to quote this exactly from Britannica: “Horsepower is the rate at which work is done. In the British Imperial System, one horsepower equals 33,000 foot-pounds of work per minute—that is, the power necessary to lift a total mass of 33,000 pounds one foot in one minute. This value was adopted by the Scottish engineer James Watt in the late 18th century, after experiments with strong dray horses, and is actually about 50 percent more than the rate that an average horse can sustain for a working day.”
Let's underscore that. As we think about the output of our engines and motors, not only are we measuring them to horses, but to ones that are significantly stronger than average. Similarly, the mile was a measurement of a presumably very fit and healthy soldier trotting at a fast clip. The poor oxen outlining a city would have to plow without resting once. It seems like so many of our measurements are based on a desire to exceed the average.
Which brings us to manpower. The term (modernized to “workforce”) reflects the total number of human workers available or required for a task, project, or within an organization. But there is a technical definition of one manpower approximately equaling 75 watts, which is about one-tenth of a horsepower. That is to say, one strapping dray horse = ten men, and one average horse = five men.
There is obviously a point where equations don’t make sense. We cannot say that all of Amazon’s work can be done by 300,000 average horses. Yet as we consider the emerging world driven by AI-juiced productivity, how should we think about the reallocation of human productivity, especially in the context of technology? Could an equation help a cost-benefit analysis? And do we have some sort of expectation, like with miles and furlongs and drays, that our metric should not just be the replacement of highly repetitive tasks, but ones that require very specialized strengths – say 1,000 Stanford computer science grads?
Ultimately, what we are trying to define - from the trek of horsepower to manpower - is the next stop on the journey: botpower. So let’s more closely consider those three questions around 1) human reallocation; 2) a technical definition; and 3) our expectations, to see what we can better understand.
Reallocation of Human Productivity
It is widely understood that AI is already automating many tasks that were once done by humans, and this trend is only going to continue in the years to come. A study by the McKinsey Global Institute found that up to 800 million jobs could be transformed due to automation by 2030, representing 15% of all jobs. This includes fields like data entry clerks, telemarketers, cashiers, tax preparers and translators.
This is perhaps undercutting it. No offense to tax preparers, for example (my mother-in-law is one), but one tax preparer doesn’t create 500 new tax accounting software programs as part of her job.
So let’s consider coders instead.
One study, conducted by researchers at the University of California, Berkeley, found that approximately 40% of the code committed to GitHub Copilot, a popular AI-powered coding assistant, was written by AI. This suggests that AI is already playing a significant role in code generation, and, profitable, with Github copilot surpassing $100M in ARR.
Another study, conducted by researchers at the University of Oxford, found that AI could potentially generate up to 80% of the code that is currently written by humans. This means that AI could have an even greater impact on code generation in the future.
Hopefully the implications of this are clear. Botpower can enable enhanced developer productivity, reduce development costs and accelerate software development cycles. Here are some examples how:
AI-powered tools can automate repetitive tasks, such as code generation, syntax checking, and error detection, freeing up developers to focus on more complex and creative aspects of programming. This could lead to a significant boost in developer productivity.
AI could help reduce software development costs by automating tasks and minimizing the need for highly skilled programmers. This could make software development more accessible to businesses of all sizes.
AI-powered tools could accelerate software development cycles by automating tasks and providing real-time feedback. This could lead to faster delivery of software products and services.
And on the human capital side, the blitzscaling method of talent acquisition goes extinct, and democratization of coding proliferates. AI could make software development more accessible to individuals with no prior programming experience. This could democratize software development, allowing more people to create their own software applications.
Of course, AI-generated code may not always be as high-quality as human-written code. Developers need to carefully review and test AI-generated code to ensure its quality and maintainability. It may introduce new vulnerabilities and security risks, and developers will need to adapt their skills and practices to work effectively with AI tools.
The Question For Enterprises
All in all, what we are facing is a massive transformation of our business building landscape. As a result, for many firms of the future, the question of “build versus buy” may easily become a relic of the past, as the default answer will always be “build.”
Why? Rather than utilizing manpower, they can increasingly utilize botpower. The repetitive stress testing, the lack of downtime, the continuous learning, the (near) total system control and the ability to utilize not just internal data but public code will enable companies to be able to build on top of existing systems and/or more efficiently produce new ones. Without the need for employees to actually know how to code, enterprises can utilize their existing workforce to create and execute complex programs without the theoretical need for third party software providers. SaaS becomes an internal function of an institution.
How should an enterprise model these considerations into their financial and business plans? Personally, I am eager to determine a precise equation to quantify botpower, inspired by Watt’s determination of horsepower. Yet maybe the cost-benefit analysis will just be pretty crude and simple: an analysis of the cost of building software internally and the number of people not hired to do it. I remember a former AI-related portfolio company of mine not too long ago that priced their product based on how many data scientists a business would not need to employ if they used their solutions.
Expectation of Greatness
At the start of this blog, we noted all the many ways that our units of measurements are based on an inflated metric of strength. As we proceed with our exploration of botpower here at Sentinel Global, we are mindful not to make too many assumptions of either the abilities or deficiencies of human productivity and the achievements or alarm generated by AI. Humility and realism must be key. We instead are focused on these primary points as we consider our investment views in AI
Skills Development: While AI may displace certain jobs, it also has the potential to create new roles and redefine existing ones. The focus should not solely be on job displacement but on identifying opportunities for innovation and entrepreneurship.
Collaboration: Rather than complete displacement, the integration of AI in coding is likely to lead to a collaborative relationship between AI systems and human coders. Successful collaboration will require coders to understand AI tools, leverage their capabilities, and contribute uniquely human skills, such as creative problem-solving, critical thinking, and domain expertise.
Quality Assurance: AI algorithms may inadvertently introduce biases, errors, or security vulnerabilities in the code. Ensuring ethical coding practices and maintaining code quality will be critical. Implementing rigorous testing procedures, conducting regular code reviews, and incorporating ethical guidelines into AI development processes are vital steps to mitigate potential issues.
Originally published at Sentinel Global.