My Current Thoughts on AI
For the last couple years, I’ve been watching the AI space in the tech industry and have been astounded by the amount of change I’ve seen. While AI has been a field of study in computer science for many years, only recently have I seen things change significantly very rapidly. And with that change, I’ve formed my own opinions on the what I’ve been seeing, and I felt it was time to at least share my current stance on it since I know that I can seem like a hypocrite between what I say, do and believe as they seem contradictory but I believe that my stance is nuanced.
Before I begin, this post and its contents are not reflective of my employer or any particular organization but rather my own personal opinions and experiences so far.
Second, I want to be clear that this is my current stance on AI and is subject to change as I learn more and things continue to change. I will try to do my best to provide subsequent updates on my thoughts and opinions, but I will try to keep this post for historical reasons as my opinions evolve over time.
Also, some of these my opinions here may already be based on outdated or incorrect information - and that should be okay. This area is changing very rapidly and my ability to stay up-to-date on capabilities, news and trends is becoming more limited as I need to focus on my own work, family and other commitments. I welcome online discourse on the subject through Mastdodon replies for this blog post, but please keep in mind that any inflamatory or offensive comments may be ignored and/or blocked. I hope that at some point I will get comments powered by Mastodon working.
Finall, while I didn’t want this blog post to be a very long read, it ended up being much more than I expected despite subdividing it into sections. However, each section is built on some thoughts from the previous one, so it sort of just grew organically as I wrote it. I’m sorry.
The Technology
Artificial Intelligence as a field of study in Computer Science has been around for decades and has experienced periodic evolutionary strides, albeit often limited to the computational capabilities of the era. In the last 2 decades, I feel that leaps in individual processor computational power have mostly stagnated and performance gains have been mostly achieved through increasing parallelization and improved algorithms. But some of the most significant computational gains I’ve seen have been through the application of mathematical theory in novel ways, allowing us to get to this point in technological history. For instance, better application of classic mathematical techniques such as linear algebra, probability theory, and Markov chains set up the foundation for modern AI systems. Through the better application of these principles, we have been able to build more efficient, effective and powerful systems.
However, I feel that the biggest catalyst to the accelerated research and development of artificial intelligence has been the open release of the research and initial applications of the training and operation of Large Language Models (LLM) and Generative AI. AI technologists open sourced their work, allowing additional researchers and developers (including those who may not have initially been fully interested or figured they could even participate in the field) to build on top of their work in an increasingly distributed and collaborative way, allowing for an exponential, global growth of the field.
While some still say that the field is in its infancy, I actually believe we are much further along in its lifecycle, exploring its applications, capabilities and limitations. I feel that there are still many, many inefficiencies, but as with any other tool, I see that it will gradually improve over time as more engineers and scientists study its problem areas, and as such has the potential to significantly augment the capabilities of human operators.
From my understanding of the foundations of these new systems, the technology is based on digital analogs of the brains of living creatures. In particular, Neural Networks share similarities to the structure and function of the brain, with neurons that accept inputs and determine outputs, whose combination determines the network’s output. Similar to the input of a stimulus to the brain, the network receives input and decomposes the information into simple signal that are then processed by the network’s neurons, where the results are aggregated and eventually an output is gathered and presented.
The brains of living creatures are highly complex, leveraging millenia of biologicial evolution to develop cellular structures that employ eletrical signals and chemical reactions to process information. These chemical reactions occur extremely quickly and are highly efficient, and when in aggregate with all of the other neurons in the brain, can input and process vast amounts of information in a very short amount of time.
In my opinion, AI datacenters are analogous to the brain of a living creature, where huge numbers of processors act as neurons, decomposing, processing and gathering resulting information. However, these systems are orders of magnitude slower than living brains, and are not nearly as efficient. As a result, these early artificial brains consume vast amounts of resources - resources that are becoming increasingly scarce.
While the endeavor to create an artificial brain is an exciting one, and one that seems to be within some reasonable reach within the remainder of my lifetime, capitalist motivations have perverted it into a race in order to fund further development of the technology at the expense of life on the planet. From a technology-standpoint alone, I feel that we are trying to push the technology for broad application too quickly. I think it would have been better to allow its use and continued research and development in a smaller sector of the industry in order for the efficiencies, capabilities and safety mechanisms to be better developed.
The Effect on the Economy
In the last couple years, for better or worse, AI’s inclusion in products and services has increased dramatically. When pressed, leaders for organizations excuse the use or inclusion of AI as critical to their business as their belief is that competitors that find ways to include it seemingly represent existential threats. In the worst of cases, its inclusion is merely an unscrupulous use of the technology in order to gain market attention and increase short-term revenue. But this feels like it is a short-sighted approach to the technology, and one that largely amplifies its inefficiencies and short-commings and may stunt its improvement over the long run.
However, the effect that has been the most concerning to me is that it is displacing human workers. I often question whether or not companies that are laying off workers due to efficiency gains as a result of AI are actually being honest. I question this because in the years prior the AI boom, companies have been laying off workers en masse due to other excuses, most of time I believe to encourage short-term funding from investors as they seek to make their quarterly or annual financial reports appear more favorably. With AI’s frenzied state in the market right now, it seems like a no-brainer to report layoffs and use the execuse of AI as a way to soften the news on decreasing headcount and attempt to shift the negative attention from the hit on humanity while at the same time trying to increase their market position as leaders in whatever industry they operate in and encourage investment income.
I personally believe that the displacement of human workers while using AI as a reason for the layoffs is creating a bigger problem: fear. Prior to the current AI frenzy, I remember that there were murmurs of an impending global rececession resulting from the draw on the economy due to the pandemic. It is well known that recession effects take some time to echo throughout the economy, and I believe that the release of AI during this time allowed for a convenient mask against the real reason. That isn’t to say that AI has not actually been a true cause for layoffs, but there are too many conflicting stories between what companies are reporting and what workers of those companies are actually saying. But I have seen (and personally am experiencing) that fear manifesting in a few ways:
- The most promenient fear I see is the fear of job loss in the face of an economy where enough jobs do not exist to compensate for the layoffs. With the effects of the recession now being felt more prominently by people, the fear of job loss is even more pronounced. As a result, I feel that many people are feeling a sense of unease and anxiety about the future of their jobs, and have resorted to “just doing what they’re told” - which includes further using AI in ways that are not necessarily useful or productive in their line of work, just because management is telling them to.
- The second thing I’ve been seeing is the fear of being made irrelevant. While it is true that AI can help make some simplier processes more efficient through automation, there are many who overvalue the capabilities of AI and as a result are prematurely optimizing out people and job functions before the AI that is meant to replace them is ready for all of that responsibility. Given my own experiences with AI, I believe it is still too early to fully utilize AI for all of the responsibilities that human workers are currently taking on, and some organizations have been realizing this as some have already started to re-hire people. But, there still does exist the future possibility that some will find their jobs functions eliminated due to AI, and this is an unease that some have felt and use that to fight against AI advancement.
It can not be ignored that the AI buzzword has caused a lot of disruption in the economy, and from my vantage point from the current use by organizations, the disruption is not positive. In the short-term, leadership will overvalue its current capabilities and stunt their own growth by overcommitting to using it, sinking enormous amounts of money on inefficient or inaccurate systems that could have otherwise been used to fund their own growth with traditional human workers. This can cause harm to the organizations in multiple ways, including diminishing margins and reputational damage. In the long-term, the overreliance on AI will lead to problems that will leave only a few options for those organizations, including either diverting even more funding to AI functions that will cut into revenue significantly enough to make the business no longer profitable, or require re-hiring human workers who likely will demand an even higher salary than if they kept those humans in the first place in order for those humans to clean up whatever mess was made.
I will not pretend to know what the future holds, but I can say that it is very uncertain where things will go from here economically. However, if I listen to the news and social media, the general sentiment is that it is not going to go up and to the right.
Social Effects
This is an area where I feel I am the least certain and qualified to discuss, but I will at least try to express what I have seen so far and where I stand. In recent years, online social media has started to be regarded as a nuisance and actually harmful. As an elder millenial, I generally agree with this sentiment, but I do disagree within a narrow window: as a tool I generally feel that social media is a powerful medium for expression, communication, collaboration and coordination, but popular platforms like Facebook, Twitter/X, Instagram, TikTok, Reddit and a multitude of others have exploited it for their own financial benefit and, through various mechanisms, have made those platforms more harmful than beneficial. Efforts to take back control of social media have been made through platforms like Mastodon, Pixelfed, Matrix, and Lemmy as these platforms try to return to the original social media roots of their equivalent popular platforms.
I see a starkly divergent opinions on AI depending on which social media platform I am looking at. In general, popular monetized platforms seem to have more posts favoring AI, while community-powered federated platforms seem to be more negative toward it. I actually seem to think that, at least for monetized platforms, there is a biased influence on the posts that I see as a) those platforms often do not show content in chronological order and promote content based on its own “algorithms,” which could be biased to up-trend AI-favorable content to satisfy financial partners (e.g. advertisers), and b) the plaforms often promote highly-contentious content in order to increase engagement time with the platform, thereby increasing the platform’s finanical revenue through paid in-feed content. On community-powered federated platforms, however, I see a predominately negative viewpoint of AI, as those platforms often have tight-knit communities of like-minded users that often echo similar opinions on AI. However, I tend to lean on the viewpoints of these users more than those on monetized platforms as I am somewhat assured that there is no influence by an “algorithm” that can either promote or silence content.
But, I think the view of these communities that AI is bad is not necessarily fair. As a technology optimist, I see AI as a tool that can be used to improve human lives and has the potential to make the world a better place, and not a tool that should be used to harm or exploit people. However, as a realist, I do acknowledge that AI can be used to harm or exploit people if not designed and deployed responsibly. And so I stand in the middle, where I see AI as a tool with limited function and requires human oversight and control, and not as a tool that organizations can use to reduce human involvement in their operations. I don’t believe AI to be the boogieman that some people have shaped it out to be, but I do see that too many organizations are being irresponsible with money and finite physical resources (such as eletricity and water) chasing after AI’s purported benefits without actually knowing if such benefits are actually possible in its current state.
Work
From a personal aspect, I am still trying to find where AI would fit in my own workflows. As a software engineer, I personally enjoy building the tools that I need in order to solve problems and improve my own life or the life of my family. The prospect of a computer taking away that hobby does not appeal to me. But, as a tool to augment my capabilities and allow me to realize the tools I want to create, AI has the potential to be a game-changer in my life, especially as a person who frequently has projects that I start but have yet to complete either due to a lack of time or energy.
However, my story at work with AI is different and very nuanced. A lot of my colleagues are both excited and afraid of the prospect of AI. Without getting into too much detail, like many organizations in the tech industry, as an R&D subdivsion we are encouraged to adopt AI and use it to accelerate our work. Though, some of us are still novices with the technology. For instance, I am still learning how to use AI to accelerate my work, but there are organizational policies that restrict my use of it, and not all of the tools that are available out there are available for us to use and as a result I am left trying to see how I can use the permitted tools in the areas where I would see the most value. These restrictions impact my ability to use different tools that may actually be better suited for my needs, and instead I spend more time trying to figure out ways to utilize the tools in the ways they were not intended.
As a technical lead for my group in the R&D organization, I try to encourage my team to learn and use the AI tools that we have been given permission to use. I figure that learning how to use these tools that are available today will not only help us at work, but also help us be more marketable in the industry should any of us ever decide to find a new career path (or, more likely the case, if we are forced to). But I do know the adoption of these tools will also yield more work and responsibility for me. In particular, I still do not feel AI coding tools are mature enough to avoid long-term maintenance issues. They are pretty good at generating code, which makes sense as tests can be used to validate the code it has generated and any failures can be fed back into the AI to continue to guide it to a solution, in much the same way that TDD allows developers to iterate on code, just faster. However, what I have not seen the AI coding tools do well is learn to write more concise code, or refactor a piece of code in order for the code to be more maintainable or reuseable across modules. This is not to say that the tools could not get there, but at the moment, I have not seen it.
This may be a short-term problem as the AI coding tools continue to evolve and improve, though it does create more work for code reviewers like myself. It was already pretty difficult to review code during pull/merge requests, but with AI generating huge swaths of code, things become increasingly difficult. Some folks have suggested that I try to use the AI tools to perform the code reviews by other people, but this genuinely feels wrong, dishonest and ethically questionable, mainly because I would be entrusting to a tool that it is certifying code that I personally would be held liable for without not actually having seen its internal construction. It would be not unlike having a building inspector inspecting the construction of a new building based on the statements of the contractors and not actually seeing how things have been put together and organized for themselves. And furthermore, if the the AI agents generated the code and another AI agent reviewed the work, what was the point of having the humans at all? This is a rabbit hole that I do not feel comfortable entering, especially as someone who works in healthcare IT.
The other issue that I have started to see happen is lack of proper review before submission. One of the stipulations I held to my team’s use of AI coding tools was that they were responsible for reviewing the code they generated and understanding it thoroughly before submitting it. If the code was not reviewed and I questioned it and they were unable to defend their submission, I would ask for it to be re-reviewed by the submitter, and I would only approve it again after they understood what it does.
There have been other areas where generated code has become a problem. For example, I have seen planning reports generated from the AI agents that were used directly to respond to design requests, with little review (if not directly submitted). This does not necessarily help me as the submissions are often extremely dense and, again, when questioned, the submitter was not able to adequately defend their design submission. So, in this early period of tool adoption and learning, despite feeling more productive, some work has actually not been effective, and the perceived productivity gains are lost when work is repeatedly returned. But, I am not going to discourage the use of AI coding tools. As a software engineering professional, I know that as usage increases, skill improves both from the tool itself as well as how the tool is used.
Closing Thoughts
The current state of Artificial Intelligence is, for me, a mixture of excitement, disappointment, apprehension and frustration. When used as a tool, I am excited to see that AI can be a productivity and capability booster, but only in the hands of those who understand how it works, how it can be used effectively, and who understand how to validate its output to ensure long-term success. Like many of the tech bubbles before it, there is a lot of disappointing fear and hype surrounding AI. I worry that companies will continue to use short-sighted strategies and try to replace humans with AI and there is a high likelihood that they will fail in the long run as a result of unrestrainted spend. And frustratingly, companies will continue to try to use AI as a marketing tool or shoehorn it into existing products without both understanding its capabilities and impact to their products and services, and without an actual problem that AI will solve in their product or service.
In fact, I believe the hype cycle may be reaching its peak as I’ve already been seeing articles along these lines. For instance, use of AI in marketing imagery is backfiring spectacularly among Gen Z consumers, and even IBM is rethinking its approach to AI. Software security is also a concern, as AI systems have been leveraged to find and exploit vulnerabilities in software (even the AI tools itself!) at an increasingly alarming rate, where information security investment has been lagging. AI has been used irresponsibly, with several accounts of AI executing actions that lead to outages and/or data loss. There is an increasing number of articles explaining how AI may atually be hurting productivity more than it is helping, and that it is also affecting critical thinking.
But I do still have hope for the technology: I have hope that the technology will become more energy and resource efficient, allowing agents to run on smaller, more energy-efficient hardware, while also improving the capabilities and accuracy of the agents to allow them to run with less and less human supervision. But I feel that we can only get to that point if we reign it in somewhat and allow niche areas in the technology industry to help develop it further for broader use later when the technology is ready. Right now, it seems very well-tuned for software development tasks, and I feel that that is a perfect area to start. But I disagree with its use in areas such as knowledge and information retrieval, and I believe should be an augmentation to regular search rather than a replacement, which is why I am a little annoyed that Google has tried to take it front-and-center in their core product rather than leave it as a secondary search alongside their traditional search engine.
I have no idea where this will go, and I don’t think anyone else does either. The leadership of companies who offer AI products and services are poor predictors as well, as there is definitely bias and incentive for them to talk up the benefits of AI while nearly completely ignoring the inefficiencies of the technology in their marketing. The only true voice I listen to are those who have been affected by it, both positively and negatively, and my opinion on the matter is formed both by their voice and my own experiences working with the technology.
Will I continue to use it? Yes, but I will also continue my pursuit to use it locally, leveraging existing resources where possible and minimizing my reliance on huge datacenters that run AI models as I do understand that they are significant draws on resources that are already limited.
