In case you’ve been living under a rock, “AI” and deep-learning algorithms have made huge advancements in the past year.
I mean, just watch ChatGPT describe itself in sailor jargon:
At first glance, this technology is terrifying. In fact, when I tried out the app Lensa this winter, which took some selfies and rendered an AI impression of me as futuristic art, I felt a deep worry in my bones.
What has happened to human creativity? What of the now unpaid artists and creators?
In addition to the abstract fear of machines cultivating art, a fearful voice in me says, “I’ll be made redundant! What about thought leadership and the special tools and processes I offer clients? Will I be replaced?”
I’m guessing you’ve had that thought, too.
But then I catch myself. Do I really need to panic?
My take? No.
In this post, I’ll share with you why I think AI and deep-learning algorithms are not something we need to fear.
Here are the three keys reasons not to panic:
- AI and deep-learning algorithms can’t do everything (and there are very specific things they can’t do).
- AI and deep learning algorithms can actually help you find the time you need to do what you’re best at.
- There is specific work that only human beings can do that AI and deep-learning algorithms cannot.
1. AI and Deep Learning Algorithms can’t do everything.
I wrote about this a few years ago in my 2019 book Bravespace Workplace.
I talked about how the rapid advancement of technology has changed the world of work, from smartphones to AI and robots.
If you want to dig into understanding where computers come from and exactly what they can and can’t do, the next section provides this deep background.
All you really need to know, though, is that every modern computer—from your high-school calculator to your smartphone or laptop or even super-computers and quantum computers—has very specific limitations which have been well-known since the 1930s.
Because of these limitations, we know that computers will never be able to do things that human beings can do, like:
- Determine whether or not a computer program will run forever.
- Determine how many steps a computer program will perform before halting.
- Figure out if different strings of characters can be recombined to equal one of the first strings.
While these examples are specific to the world of computer science, by understanding them, we can begin to see the contours of what humans are best at.
For a full list of what I argue is human’s highest and best use at work, see below.
A quick review of the history of computer science, with a lens towards the development of modern computers and their limitations.
For this in-depth guide into the limits of so-called “AI” and deep-learning algorithms, I want to hone in on the artificial intelligence part of tech by going back—way back to the year 1900.
We’re in Paris at the International Congress of Mathematicians.
It’s only the second time this important group of mathematicians has met, and the first time they awarded the Fields Medal, honoring mathematical contributions by people under 40.
You’ve got to picture one of the nerdiest events of the day:
At one of the primary presentations of that conference, David Hilbert proposed 23 problems that he deemed to be the most important problems for mathematicians of the 20th century. It was a strong call to action for mathematicians to focus on solving these problems and advancing the field of mathematics.
Attempts to solve these problems in the ensuing decades laid the theoretical groundwork for all of computer science today, including AI and deep-learning algorithms like ChatGPT, Dall-E, and others.
Specifically, we’re interested in what’s called “The Entscheidungsproblem.”
The Entsheidungsproblem – what can machines do?
We’ll get a bit technical in describing the Entsheidungsproblem, but it’s important to understand the limits of modern computation.
The Entsheidungsproblem, also known as the Decision Problem, was the problem of finding an algorithmic method to determine whether any given mathematical statement or proposition could be proven or disproven within a formal axiomatic system.
In other words, The Entsheidungsumproblem asks is there an algorithmic way to determine if there’s a possible solution for this (any) problem. Some mathematical problems required so much work mathematicians wanted to know if there was a way to know whether or not a given problem even had a solution before jumping into the laborious of finding that solution.
Hilbert believed that the solution to this problem would have far-reaching implications for the foundations of mathematics and logic.
In the years that followed Hilbert’s lecture, mathematicians and logicians worked on the Entscheidungsproblem, leading to the development of the field of mathematical logic—the sets of proofs and theorems that underly basic arithmetic—and the eventual discovery by Kurt Gödel in 1931 that there are mathematical statements that are undecidable, meaning that they cannot be proven or disproven within a formal axiomatic system.
The discovery of undecidable statements had significant implications for mathematics and logic and led to a better understanding of the limits of what can be proven using formal systems.
Enter “The Turing Machine”
Here’s where we get into the juicy under-belly of all modern computation, with Alan Turing’s 1937 paper On Computable Numbers, with an Application to the Entscheidungsproblem.
This paper is one of the most important papers in the history of computer science. In it, Turing introduced the concept of a universal computing machine, which is now known as a Turing machine.
A universal computing machine was, in essence, the theoretical model of binary logic upon which all modern processors function.
It’s this paper in which Turing describes his “Turing machine,” a theoretical model of a computer that can perform any computation that any other computer can do. It’s a computer on a computer, so in one stroke, Turing invented modern computers and proved the limits they inherently possess.
This paper had wide-ranging implications, but in the context of the limits of modern AI and deep-learning algorithms, it described the notion of computability, which is the ability to solve a problem using a mechanical procedure.
Other implications included the following:
- First, it showed that some problems are not solvable by a mechanical procedure. This led to the development of the theory of computational complexity, which is concerned with studying the inherent difficulty of algorithmic problems.
- Second, Turing’s work laid the foundation for the development of modern computing. The Turing machine is a fundamental concept in the theory of computation, and many modern programming languages and computer architectures are based on the principles that Turing laid out.
- Finally, Turing’s work had important implications for the philosophy of mind. Turing’s ideas about computation and computability helped to inspire the development of artificial intelligence, and his work has had a profound impact on how we think about the nature of human thought and consciousness.
All of this leads us to the concept of “computability,” which clearly outlines the fundamental limitations of modern “AI.”
Computability – here’s what “AI” & deep learning algorithms can and can’t do
Computability is a fundamental concept in the theory of computation, and it has important implications for artificial intelligence and deep-learning algorithms like ChatGPT, Dall-E, and others.
At its core, computability is the ability to solve a problem using a mechanical procedure, such as an algorithm or a computer program. This means that if a problem is computable, there exists a mechanical procedure (read: algorithm or computer) that can be used to solve it.
However, some problems are not computable and cannot be solved by any mechanical procedure, including AI and deep-learning algorithms. For example, the Halting Problem, which asks whether a given program will eventually halt or run forever, is not computable.
In practice, AI and deep-learning algorithms like ChatGPT are designed to solve specific types of problems, such as natural language processing or image recognition. These problems are generally computable, and the algorithms are designed to find solutions using various mathematical and statistical techniques.
However, it is important to recognize that AI and deep-learning algorithms have limitations. They are not capable of solving all problems, and there may be some problems that are beyond the scope of what these algorithms can handle.
If you want to get specific about what deep-learning algorithms can and can’t do, I’ve provided a list of some non-computable problems and how humans might find solutions themselves.
A list of some non-computable problems and their application to daily life.
|Non-computable Problem||Description||Why it’s non-computable.||How it’s relevant to everyday life||How a human might solve this problem|
|The Halting Problem||This problem asks whether a given program will halt (i.e., stop running) or run forever.||It is impossible to write a program that can solve this problem for all possible inputs.||This problem relates to computer programming and software engineering, which are used extensively daily. The halting problem arises because it is impossible to know whether a particular program will stop running or continue indefinitely. Only humans can discern if this type of error will lead to bugs and errors that affect software performance and can cause problems for users. This is highly relevant to any automatic computer processes, such as those used in self-driving cars or in-home security systems.||A human might attempt to solve this problem by analyzing the program and determining whether it contains any infinite loops or other constructs that could cause it to run indefinitely.|
|The Busy Beaver Problem||This problem asks for the maximum number of steps that a Turing machine with a given number of states can perform before halting.||It is impossible to compute this value for all possible inputs.||This problem is related to the study of computation and is important in the field of computer science. It is used, for example, in the design of algorithms and in the analysis of computational complexity. In everyday life, this is relevant to the development of technology and to the advancement of science, such as the use of robots in highly complex environments, such as tornadoes, that have limits as to what is computable.||A human might attempt to solve this problem by enumerating all possible Turing machines with a given number of states and then simulating each machine to determine the maximum number of steps it can perform before halting. However, this approach is impractical for larger numbers of states, and it is not possible to compute this value for all possible inputs.|
|The Post Correspondence Problem||This problem asks whether there exists a sequence of strings that can be combined in different orders to produce two equal strings.||It is impossible to write a program that can solve this problem for all possible inputs.||This problem relates to the study of formal languages and automata theory, which are used in computer science and linguistics. It is used, for example, in the analysis of natural language processing (e.g. ChatGPT or Dall-E) and in the design of programming languages. While not applicable to most of everyday life, it’s worth observing that humans are quite good at solving this type problem. It’s tedious, but even grade-school children could look at two strings and solve to find if there’s a way to repeat them (think of an anagram puzzle). This has implications for the constraints of things like mutations in DNA sequencing.||A human might attempt to solve this problem by trying different sequences of strings and combining them in different orders to see if they produce two equal strings.|
|The Entscheidungsproblem||This problem asks whether a given mathematical statement is provable or not.||Turing’s work showed that there is no general algorithm that can solve this problem for all mathematical statements.||This problem is related to the study of logic and mathematics, which are used in everyday life in a wide range of contexts, from financial transactions to scientific research. The problem of deciding whether a mathematical statement is provable or not is relevant to the study of logic and to the development of mathematical theories.||A human might attempt to solve this problem by using logical rules and deduction to determine whether a mathematical statement is provable or not. However, Turing’s work showed that there is no general algorithm that can solve this problem for all mathematical statements.|
2. AI and Deep learning algorithms can give you more time to do what you’re best at.
Back when I wrote Bravespace Workplace in 2019, I started the chapter on AI with this quote from Isaac Asimov—and I think it’s even more true today than ever:
In a properly automated and educated world, then, machines may prove to be the true humanizing influence. It may be that machines will do the work that makes life possible and that human beings will do all the other things that make life pleasant and worthwhile.
— Isaac Asimov, Robot Visions
Asimov claims that there’s a way to use machines that make our life even more human.
With an example like ChatGPT, the use case is pretty obvious.
Use ChatGPT to make the boring, tedious, or distracting work you do easier to have more time to do more important work. See #3 below for more on what that “more important” work is.
And if you want a cheat sheet of 10 ways for lazy managers and leaders to use ChatGPT, download it below.
10 Ways for Lazy Managers & Leaders to Use ChatGPT
3. There is specific work that only human beings can do, and AI and deep-learning algorithms cannot.
The work that is our highest and best use as human beings is not computable. In fact, it is what only people can do. So what is our unique work as humans?
It’s everything we talk about in People Leadership, things like:
- Solution Discovery Coaching
- The Heart Habit
- Resolving conflicts
- Using courage to do hard things
- Building connections in teams
- Growing an unbreakable culture
- Developing leaders who are good for people
- Repairing broken trust
- Flexing how we show up for human differences
- And so much more!
In short, the uniquely human capacity to feel with, work with and support human beings.
And fun fact: these very things are what consistently, relentlessly, and tirelessly occupy vast amounts of time for people leaders at every level.
How effective would organizations be if the humans in them could focus on these things instead of the tasks that AI and deep learning could make easier and more efficient?
How will you use the newest generation of AI and deep-learning algorithms?
I’m curious how you already are or envision you or your company using the newest generation of AI tools and deep-learning powered software.
My team’s already using ChatGPT to make brainstorming easier, to summarize important articles, or to help write blog posts like this one.