The flip side of generative AI: Extractive AI
No need to worry about plagiarism. Responses, which are often exact quotes, are more structured, consistent, and precise, accompanied by traceable crumb trails back to the sources, and include an explanation of the logic or reasoning that was applied. Extractive AI can also assign confidence intervals and other metrics to the outputs. As opposed to GenAI, which is more like a closed-book exam, extractive AI can be viewed as being of the open-book variety. Now let’s address the elephant in the room when it comes not only to GenAI but also to the eventual emergence of AGI (artificial general intelligence). That is, the fear of widespread job displacement.
KM to the rescue
One consistent thread in this column regarding future tech trends has been an increased, as opposed to minimized, role for humans. That role always has and will continue to be centered on KM. Critical human functions include assessing veracity, sense-making, and the incorporation of intuition, expert judgment, and wisdom. Not to mention that powerful economic engine known as innovation.
Despite achieving increasingly humanlike behavior, computers will always be computational. Their sole function is crunching zeros and ones. Their outputs are purely analytical, based on numeric weights and scores with no understanding, emotion, intuition, or, above all, any sense of stewardship. Equally important is the notion of novelty. True, an AI system can come up with a new idea or concept, even a new theory. However, it will still take a human to determine whether or not an idea makes sense or can be considered ethical. And don’t forget the occasional black swan event, which rarely shows up in LLMs.
Any semblance of intelligence arises mainly as a behavioral trait, such as the ability of the system to closely approximate human conversation. In that sense, machines learn in much the same way that children, and even adults, learn: through imitation. They’ll repeat what they hear, even though they may not fully understand what they are saying. In many ways, “monkey see, monkey do” has become “AI see, AI do.” The difference is that AI has access to massive volumes of existing data, but often with little insight into how the answers are generated or confidence in their veracity.
There are also signs that the supply of data might be reaching its maximum, with progressively smaller amounts of data remaining to be “mined.” This leads to the question, “Then what?” The answer should be obvious. More than 8 billion minds on the planet are overflowing with every type of knowledge imaginable, including untapped Indigenous approaches to science, mathematics, health, learning, and, perhaps best of all, different ways of simply looking at the world.
A perfect match
Clearly, there’s a place for each of these two types of AI, with a crucial role for KM in each. On the generative side, we have sense-making and prompt engineering to refine not only the queries, but also the underlying machine learning models and algorithms. On the extractive side, it’s building that much-needed body of knowledge regarding what blend of the many available AI modalities works best in given situations, and the extensive curation and governance that goes along with it.
Want to carry on an interesting, intelligent conversation about any subject, on demand, 24/7? GenAI is waiting. Need a precise, validated answer to a question or problem with serious cost, legal, or other implications? Then extractive AI is what you’re looking for.
Yes, GenAI is exciting and will continue to evolve. But let’s not become so enamored with it that we miss out on the many opportunities available on the extractive side. Better yet, let’s put our collective minds together and find ways to make the most of both of these brave new worlds.