Machine Learning: Present and Future

    Posted: Jul 11, 2023

    Robots and Human Concerns

    Concerning artificial intelligence, two distinct fears haunt the dreams of science fiction writers and their audiences. Firstly, at some point, it will become impossible to know the difference between an authentic human and artificial intelligence. Secondly, humans will no longer work in the future and will therefore be irrelevant. Both fears are at work in our contemporary discussions of labor and artificial intelligence. But are these fears justified?

    Honda’s ASIMO notwithstanding, it is not generally useful to fashion a robot into human form since we already have many humans doing the kind of work that humans can do. In the same way, as Amit Shkolnik shows, machine learning does not mimic human intelligence, nor should it. We need machine learning to formulate the insights that human intelligence finds difficult to provide. We need not fear that robots will be indistinguishable from humans, because there is little purpose to fashioning robots and artificial intelligence after human models, except perhaps, as the poets imagine, for deception. Lacking the innate social and emotional intelligence of humans, machines tirelessly perform calculations that many people might be able to do, but the machines do these much faster and with more consistency. “Computer”, after all, was first a job title.

    The Anatomy of Intelligence

    While it is fun to consider the anatomical features engineers may give to robots, the more immediate concerns have to do with the anatomy of intelligence. A self-driving car, for instance, does not look very different from any other car. The external anatomy is already there. The intelligence is not yet ready.

    As experts in end-to-end data platforms, Ultra Tendency builds the sort of data infrastructures consisting of both edge computing and central analytics that are the foundation of the artificial intelligence revolution. It is not discreet physical entities that must learn and act. With machine learning, a virtual entity can learn. A factory can learn. A business process can learn. These are not robots in the usual sense and you are less likely to find exciting videos about these learning entities, but these, more often than individual robots, are at the center of the AI revolution. The reason is that in an Internet of Things, a robot is just one more thing and by itself will not have all the memory and processing power of an entire system.

    In one example of an IoT Predictive Analytics use case, Ultra Tendency created a smart factory with the following components.

    1. Raspberry Pi edge devices collect power level and temperature data.

    2. An Azure Event Hub ingests the data from Raspberry devices via standard TCP. That means the data from various sources is stored in a central medium where it can be accessed.

    3. Spark on a Kubernetes cluster accesses and processes data, identifying anomalies, with minimal latency.

    4. Processed data and anomalies that have been identified are stored in an Azure Redis cache to feed into a user interface where data scientists can access anomaly data.

    5. Alternately, unprocessed data from the Event Hub goes to Azure Data Lake Storage (ADLS)

    6. From the data lake, data scientists train models, run batch queries, and publish the data back to Spark on Kubernetes in a Docker container, so that Spark can utilize trained machine learning models to better identify anomalies.

    The insights gained from this anatomy are not necessarily automatic. It still requires the use of data scientists to produce and interpret insights. Anomaly detection in industrial processes is, however, a product of non-human intelligence. No human could make calculations in the necessary time as Spark does. A basic principle of big data is the distribution of resources. ADLS allows for the distribution of data storage across multiple clusters while Spark likewise allows for the processing of data in a distributed manner. So the machine learning trained pipeline can learn from the data, identify anomalies, and make predictions.

    Artifical Intelligence vs. Human Intelligence

    While the human brain is arguably also a distributed system, Industrial IoT machine learning will not be mistaken for human intelligence, nor is there any reason to believe that AI will ever resemble human intelligence because its jobs are non-human in nature. As of now, this kind of machine learning pipeline helps industrial processes become more efficient and therefore reduces the need for human-sourced maintenance. This can reduce human employment on the margins, freeing workers to perform more fulfilling tasks and enjoy work and life more fully. Just as the intelligence of human laborers is wasted on repetitive physical labor, the intelligence of knowledge workers is wasted on labor that can be completed by a machine.

    Already, lawyers use machine learning to identify similar cases and precedents across the vast history of legal decisions. It is no longer necessary to have clerks poring over law tomes, searching for precedents. But the AI cannot yet make a series of cogent legal arguments. While this reduces the need for employment in legal professions it also frees lawyers from the drudgery of that kind of labor. On the whole, humans are better off when freed from boring drudgery.