Synthetic Intelligence
Carl Olofson, DBMSGuru LLC
Independent Analyst | Recognized Thought Leader | Market Research
AI is the current darling of the IT world. Some people expect that it will eliminate millions of jobs and make the remaining jobs far easier and more productive. Perhaps. But we must also be realistic about what AI can do for us. The term seems to imply actual intelligence that is produced artificially, rather than by humans. In this document, I suggest that it is really a facilitator of intelligence and has some of the external characteristics of intelligent behavior but may be better labeled “synthetic intelligence”.
The term AI (artificial intelligence) can refer to any of several functionally related technologies that perform actions that require the involvement of some level of intelligence. Some of these technologies have existed for decades. They have been implemented using decisioning algorithms and others deployed using neural networks. Currently, the most popular AI technology in common use is generative AI (GenAI), which has been made practical by applying huge amounts of processing power (enabled by graphical processing units, or GPUs) against massive amounts of source data. This can be used to respond to any human language prompt with a human language answer based on the processing of massive amounts of textual data, or to generate graphics based on the processing of massive amounts of graphical data.
GenAI, however, is not really intelligent in the conventional sense. It doesn’t “think”. It doesn’t reason, analyze, or make judgments based on any logical process. Rather, GenAI works by correlating massive amounts of data (usually text) and building a structure that enables data components (or chunks) to be organized by a proximity model based on the patterns of correlations of those chunks in the source data. The resulting structure is called a large language model (LLM). When a request comes in (in the form of a prompt), the GenAI service feeds the prompt elements into the LLM, which performs a scan of chunks that are in close proximity to those from the prompt and uses predictive analysis algorithms to put the relevant chunks together. It then uses a natural language process to render a response in human language.
Usually, the response is fairly useful. Sometimes it is a little off. Sometimes it is wildly off, returning what are commonly called “hallucinations”. Various techniques have been developed to keep the GenAI result within the contextual bounds of the prompt. Such techniques include guiding constraints (such as retrieval augmented generation, or RAG) and coded logic modules (agents) that keep the process on track. These can greatly improve the quality and relevance of the response.
Since the GenAI operation is not based on elements of what we usually call “intelligence”, I propose that it should be called “synthetic” (or simulated) intelligence. This is more of a pattern-driven calculator (albeit on a massive scale) than a realization of a Star Trek style artificial life form (e.g., the android Data). It can save massive amounts of effort in doing research and drafting summaries of relevant facts, and because of the scope of data involved, can generate far more complete results than a human is capable of. It is highly useful in providing a first draft of research results and of course can be used to generate routine texts such as press releases and meeting summaries. But we should be aware of its limitations.
Wild claims have been made recently that GenAI will replace humans in most areas of endeavor, making better choices faster, writing better and faster, and exceeding human capacity in many areas. While it is right to say that GenAI can save knowledge workers a tremendous amount of time and can enable them to research massive amounts of information very quickly, it does not replace human creativity, insight, or imagination. GenAI does not actually create anything; it simply reorders existing information and generates combinations based on already existing patterns.
Recently, Microsoft announced a 3% reduction in force, flattening its organization structure and increasing the number of direct reports to managers, all on the basis of the idea that with AI technology, the volume of concurrent activities that managers can oversee is greatly boosted. Microsoft isn’t the only organization making moves like these. But effective management isn’t just tracking the progress of projects and the productivity of the team. It also involves human interaction, including mentoring and providing guidance and encouragement. I’m not sure AI greatly expands the capacity of managers to do those things.
AI in general, and GenAI in particular, are great tools that can profoundly expand our ability to do a wide range of jobs, but it may be too soon to make radical changes before we really see how this technology impacts business processes and organizations in practice.