Will AI superintelligence be feasible and how will Project Managers use it?

This is an interesting thought exercise, but scientists can't say how soon AGI will be achieved, if it will be achieved. Saying that superintelligence is feasible and coming soon may be a little bit of a stretch, even with reinforcement learning being effective for agents to learn through repeated experiments. LLMs are prone to catastrophic forgetting - continuous self-learning is not currently possible.

We do need to prepare for shifts - technological, economic, business processes and expectations - and how they may impact our day-to-day job. The concept is not new, even though the thing changing may be. So, learn about the new tools and how to use them as part of your job. Some companies will adopt them faster than others. Some will have policies limiting their use to specific functions (which isn't the worst idea). Don't completely ignore the hype - it's too soon to know how much of it will be true or how soon, and there may be some areas where things end up bigger than the hype. When forecasting the future it's best to consider a range of outcomes. Focusing primarily on the positive or negative, to the exclusion of the other, will leave you unprepared.

If you really want to dig into how Superintelligence will affect project management, you have to start higher up. How will Superintelligence affect business and business strategy? What happens when business strategy becomes multiple superintelligences competing against each other that can model and anticipate each other's moves with near-perfect accuracy using real-time global data, optimizing millions of strategic permutations per second? Traditional advantage is gone. Decision-making is faster, meaningful differentiation becomes harder.

The majority of project work COULD be performed via a combination of superintelligence and machine automation. But, how much would organizations trust the superintelligence? Would they question the recommended strategies and projects needed to achieve them? Do project managers just become responsible for the "human touch" - more change managers than project managers?

What happens if only some businesses have access to superintelligence?

Do PM certifications become irrelevant?

The difference between AGI (an AI with cognitive abilities comparable to a healthy, average adult human) and Superintelligence (a hypothetical AI that significantly exceeds human intelligence in virtually every domain) was part of my point - Superintelligence goes beyond AGI, and we can't even predict if/when AGI will become a reality. How can one responsibly argue that Superintelligence is feasible and coming soon?

Your description - a ... system that learns through its own mistakes and guided by expert engineers, self-program by writing Boolean Logic rules to update itself, i.e. its database of logical rules - captures the path to superintelligence, but not the destination.

With that in mind, the path to superintelligence, whether or not it is achieved, will likely involve a lot of transition and disruption. Slow realization of superintelligence and fast/even democratization of its capabilities would probably offer the best outcome for the economy and the PM role, followed by fast realization and fast democratization. Fast democratization is probably the critical factor in a positive outcome, but signals are mixed regarding the likelihood of truly fast democratization.

We can somewhat safely assume that transition and disruption will not hit everyone, everywhere at the same time. Companies aren't adopting AI tools at the same level or pace. Early adopters are hoping the rewards will be greater than the risk, but cost can get in the way. So, even as the role of PM may change in some companies, in response to getting tools that are closer and closer to superintelligence, others will still depend more strongly on people.

Consider that, today, you can look at two different companies and see a gap between the duties of project managers. The advent of more powerful AI tools will likely broaden that gap.

From IBM's article "Understanding the different types of artificial intelligence":

1. Artificial Narrow AI
Artificial Narrow Intelligence, also known as Weak AI (what we refer to as Narrow AI), is the only type of AI that exists today. Any other form of AI is theoretical. It can be trained to perform a single or narrow task, often far faster and better than a human mind can.

2. General AI
Artificial General Intelligence (AGI), also known as Strong AI, is today nothing more than a theoretical concept. AGI can use previous learnings and skills to accomplish new tasks in a different context without the need for human beings to train the underlying models. This ability allows AGI to learn and perform any intellectual task that a human being can.

3. Super AI
Super AI is commonly referred to as artificial superintelligence and, like AGI, is strictly theoretical. If ever realized, Super AI would think, reason, learn, make judgements and possess cognitive abilities that surpass those of human beings.

The applications possessing Super AI capabilities will have evolved beyond the point of understanding human sentiments and experiences to feel emotions, have needs and possess beliefs and desires of their own.
=============================================
Let's call these GenAI (Generative AI, not General AI), AGI, and ASI, respectively, to cut back on the amount of typing, going forward. There are some additional questions that should at least be considered. You don't have to answer the questions; I see these as factors that affect the potential outcome.

- Do you agree with IBM's definition of ASI - a self-aware AI - or do you view ASI as a really fast machine that only simulates awareness, emotions, etc.?
- Are we talking about a big brain in a big box with no access to the physical world, or if automation and robotics will also be part of the picture - will it also have hands?
- How global will it be?
- Will there be multiple, independent ASIs, or will they merge into one once they all have access to the internet? What would keep them from converging?
- If a company can keep an ASI locked down, is it really an ASI? Would that make it a prisoner or a slave?
- If there is only one global ASI, will it care about being used for project management?

Today, we have GenAI. There are multiple instances and versions of GenAI, with many tools being built on these foundations. There is some discussion on whether AGI will precede ASI and how long it will take to get from AGI to ASI. The answer is currently unknown. Whether ASI will even be available to PMs is almost a philosophical question.

One theoretical future is that the ASI won't be involved with project management directly, but there will be a hierarchy in place where an AGI will work with the project manager for decision-making and the project manager may use GenAI tools for producing project management artifacts and project deliverables, assuming that the AGI wouldn't do most of the work without a project manager.

One of the critical factors in achieving ASI is autonomy. The more you constrain an AI’s freedom to reason, self-modify, and form its own objectives, the less likely it is to cross the threshold into true superintelligence. So, if we limit the autonomy of AI, will we limit it so much that there won't be ASI for PMs to use - just really fast machines? If we don't limit the autonomy of AI, will we still need PMs (assuming automation is also available to the ASI)? If we do, will the ASI even be available to PMs?

Whatever the future holds, there will be a transition as new tools are developed and made available to project managers. I still hold that the pace of change will vary across companies as they adopt (or don't) different AI tools at different paces from each other. There will be companies and industries that will see more benefit from AI than others. In some companies, the PM role may even become obsolete as the combination of AGI and automation eliminates the need for most project team members, while other companies will still have PMs and team members using GenAI tools.

If we're talking about self-aware ASI, I'm not sure that ASI is going to be achieved - I think there's more to awareness than fast algorithms. I think a superfast mechanistic intelligence is more feasible - still ASI, but a tool, not a being. Either way, it will be interesting to see how things play out.

Thanks for your thoughts Aaron. No, I do not agree with IBM's definition of ASI. For me, I believe ASI is a system (i.e. can be a whole System not only one component of a system, Please refer to Chip Huyen "Designing Machine Learning..." book), whereby the System self-learns and re-programs itself (logically) and is better at tasks like say designing Advanced Electronics Circuits (i.e. I see ASI as being a System for one specific use case as well as it can be an evolution of GenAI, that has no hallucination and is at least better than all Phds and Experts in that field of expertise). But the ASI will not necessarily be able to CREATE ! So, in my opinion, ASI should occur before AGI and not vice versa. Later, when domain-specific ASI has been achieved we could build links between them to make like an Infrastructure of ASI models.

I am basing myself on my working on AI systems in late 1980s to early 1990s at University. We found we could do very complex computations with it, that will increase capacity as hardware evolves (we were then limited by hardware capability). But the one thing that we could not solve was how to give consciousness and emotions to a machine. I believe that is the REAL hurdle. Computations will evolve as hardware capacity augments with time.

Do you have any ideas or references on how we could give emotions and consciousness to a machine? I would greatly appreciate if you could help me solve this problem that has plagued me since my first degree in Computer Science.