What AI Really is and Why Businesses Keep Getting It Wrong
- 3 days ago
- 7 min read
Kass James is an assistive technology specialist with a master’s in management of information systems from the University of Houston’s Bauer College of Business. Fully licensed in ADA compliance and environmental access, he’s a partner at The Spoonie Advocate Associates.
It has become impossible to discuss the future of business or IT project roadmaps without a serious conversation about artificial intelligence. But what do people mean when they are talking about AI? Schools, governments, and businesses discuss how to manage AI systems, yet they seem to know nothing about them. They talk about AI as though it were science fiction. Every website, lawmaker, tech billionaire, and IT professional seems to have a different opinion. When we talk about the cost and value of artificial intelligence, it often seems as though many of us are talking about different things.
It is deeply frustrating for those of us in the IT industry when C-suite executives discuss AI without understanding it. It is going to upset a lot of people, but we must start with the fact that artificial general intelligence, known as AGI or “true artificial intelligence,” does not exist yet. Many things seem like AGI. They mimic it and are even claimed to be close to it, but they are not.
Artificial general intelligence refers to a computer’s ability to make independent decisions, draw conclusions based on feelings rather than facts, adapt and change opinions fluidly, and approach “human” thought. We do not have it, and despite the claims of some incredibly wealthy people, we are not even close to cracking it.
We are limited by current technology. Anyone who is trying to sell you or your company a “replacement” for your human workforce is selling you the pipe dream of AGI. It is, however, a goal of many companies to develop the next best thing.
If AGI does not exist, what do they mean when they are selling AI?
There are a couple of things that fall under the umbrella of artificial intelligence, but usually, what is being discussed is a machine learning, or ML, algorithm. ML is the basis for every single AI system being pushed, and it is the only way we have to simulate intelligent decision-making. It is not actual decision-making, but rather a combination of data collection and a well-trained predictive algorithm.
IT professionals train machine learning algorithms through data and repetition
The system starts with data collection. The more extensive and up-to-date the data, the better. This is where many legal objections have been raised, as customer privacy remains a legal gray area. Common buzzwords such as “data mining,” “web mining,” and “scraping” all refer to the process of collecting large amounts of data. This is also what those large data centers are designed to store. The data is then sorted, categorized, and tagged with a range of characteristics. The algorithm is initially trained manually through supervised modeling and learning, allowing it to learn to “cluster,” or group, similar items.
It then narrows its focus and excludes anomalies. Finally, it begins making structured predictions based on interactions with the data and users. These are not actual predictions. The algorithm uses statistical probability to produce results based on data points that often appear together. This is most often seen in systems such as YouTube and Netflix, where the algorithm makes recommendations based on the videos other users have watched. It does not, however, care whether the results are correct.
AI algorithms are addicted to interaction, and you are their dealer
The downside of how ML is developed and trained is that the systems only function if they focus on what drives interaction. This means the algorithm does not know or care whether the results are accurate, only that users interact with it often. This is the failure point for many of these systems.
They will generate results that appear authentic based on the search terms if no results are found in the database. This is a problem for both businesses and legislators because these results are often mistaken for authentic information and have contributed to widespread misinformation.
The reason for and burden of data centers
Data centers are a necessary evil in this industry. The data collected and the computer infrastructure used to interpret it are massive and constantly growing. Nearly 400 terabytes are added to the internet every second, with an annual growth rate of 15 to 25%. The infrastructure used to gather and interpret the data also needs to scale accordingly. Unfortunately, the costs associated with these centers are not just monetary. They also include demands on infrastructure and natural resources.
Water and power are necessary for cooling and operations. However, data centers seldom shoulder the burden of upgrading local infrastructure to meet these needs. In addition, data centers produce very few permanent jobs and contribute little to the local economy. Thus, it is understandable that communities oppose building projects in which local taxpayers shoulder the burden of infrastructure upgrades while seeing no long-term benefits. Data centers are necessary, but developers seem unaware that, if they want to build, they must budget for the necessary infrastructure.
What is not AI or machine learning?
Mistaking complex software for AI seems to be a common problem in education, where generative AI is understandably prohibited. Still, administrators do not realize that not all software designed to assist users is AI. Text-to-speech and speech-to-text systems may be used by AI systems to collect data and interact with users, but they are not generative AI.
Similarly, basic spell-checking and grammar-checking software is not AI, but more advanced systems, such as Grammarly, may use AI to edit. Business systems such as basic chatbots, automated phone menus, and statistical analysis software are also not AI.
So, where are ML systems used?
The most common examples are search engines and data interpretation software systems. This is also the best use of AI. You ask a question and get results. These systems excel at the inhuman task of taking in trillions of data points, sorting them, and producing results based on statistical predictions. Brokerage firms have spent sums equivalent to the annual GDP of many first-world nations on these predictive analytics systems, and for good reason. Google has become synonymous with searching online by developing the best search algorithm money can buy. You also see the predictive side of these search engines in the recommendations from systems such as Netflix and YouTube.
Targeted advertising is the second most common interaction you will have with machine learning. For better or worse, you are nothing but a dataset to a machine learning system, and the more it knows about you, the better it can try to sell you something. On the other side of that, digital marketing firms pay a premium to know what drives interaction and which advertisements are not engaging. Google’s and Facebook’s market dominance lies not in social networking but in targeted advertising and data mining. The more people interact with something, the happier that system is, and the more profitable it can be for an advertiser.
Generative pretrained transformers, or GPTs, have become somewhat synonymous with AI, but they use all that data to generate a response that the algorithm believes will drive interaction. Systems such as chatbots and AI assistants can be extremely useful for providing immediate service to customers without needing to scale your customer support system. The problem with these systems is that falsehoods, errors, and “hallucinations” become more prevalent the more they are relied upon to generate new and unique data.
Because they are trained to provide information based on likely interaction, these systems often create responses that look true but are incorrect assemblies of information. One of the most common examples occurs when professionals search for peer-reviewed research, and the GPT system provides the title of a paper that was never written, attributed to authors who have never worked together. Once again, the system is an addict and will do anything, including assembling bits of information to fabricate a result that looks like what you asked for, to get that interaction.
Where AI really works, and we do not talk about this enough, is in making people more efficient. New video cards use AI to process data more efficiently. Data modeling and predictive systems for weather, natural resources, and healthcare reduce human error and save lives by predicting change. AI is also necessary for many programming languages, such as Python. Correctly applying AI can reduce errors and employee burnout while increasing workforce productivity.
The value of AI lies in how you use it
It is important to remember that AI is not inherently good or bad. It depends on how it is trusted and applied. ChatGPT is not your friend or personal assistant, but using it to gain a new perspective can make you a more effective leader. The best AI tools for programmers do not replace them, but AI-assisted development can make your team less prone to errors.
Chatbots streamline your customer support system, but you will always need a human to tackle complex issues. Healthcare systems such as AlphaFold 3 allow us to fully sequence and model DNA in days rather than months or years, yet scientists are still needed to interpret and use that information. Effective use of AI can make a company more effective, but relying on these systems or using them to replace humans completely is where everything falls apart.
Read more from Kass James
Kass James, Healthcare Business and Disability Specialist
Kass James is a forerunner in the field of disability rights, corporate responsibility, and healthcare business. Having been physically disabled for most of his life, Kass was acutely aware of the lack of accessibility in the workplace. His work focuses on restructuring healthcare to increase profitability while benefiting patients, as well as doing patient assessments for ADA compliance and assistive technology. He’s a partner with the Spoonie Advocate Associates, an organization pushing for increasing value and patient outcomes through common sense and responsible change.
