TECH4GOOD
The debate continues about the yin and yang of artificial intelligence. As mentioned in previous Tech4GOOD columns, this is just the beginning as the technology itself continues to evolve. But how many of us realize that the technology itself is a product created with the use of human intelligence? That human intelligence is actually the basis of artificial intelligence?
The creation of AI systems like ChatGPT is a product of a tedious flow of work done by thousands of scientists, data engineers, data annotators/labelers, and researchers who may be working independently from each other from all parts of the globe. They are continuously developing new algorithms, and techniques, and preparing learning data to improve the capabilities and reliability of AI systems. Human intelligence is therefore responsible for designing the systems, developing the software referred to as algorithms, and training them on large data sets that would allow them to interact with the world intelligently.
Developing an AI system requires a combination of technical expertise, analytical and problem-solving skills, tools, and research skills. It is a complex process that requires expertise in new domains such as natural language processing and machine learning in addition to security, privacy, and computer vision skills. All of these are used in every step of the process which would usually include defining the problem that the AI system intends to solve, collecting and preprocessing data needed for the development of the models, training and assessing the models for accuracy and reliability, deploying the models, and integrating them with other systems.
The human intelligence behind artificial intelligence is constantly evolving as new research is conducted and new algorithms are created. As AI systems become more sophisticated, they will require more sophisticated human intelligence to design, develop, train, and deploy.
One critical area where human intelligence is used is in data annotation which is the tedious work of converting raw data sets into the right and contextual data used to train AI models. The AI system learns by finding patterns in vast quantities of data that has to be sorted and tagged first by human data annotators/labelers, a vast workforce usually anonymous and scattered all over the globe. An example is the labeling of video footages used for training self-driving cars. It has to be done properly otherwise it can result in creating havoc out in the streets. It’s difficult and repetitive work but very critical in ensuring that the vehicles will be traveling the streets safely. Some years back, a self-driving test car operated by a popular ride-sharing company killed a pedestrian because, though it was trained to avoid other vehicles, road barriers, cyclists, and pedestrians, it didn’t know what to make of someone walking a bike across the street.
As mentioned, human intelligence is vital in the design and development of algorithms and models which are the building blocks of AI systems. Their design and development are complex and challenging tasks but are essential for the development of powerful AI systems. Algorithms are classified into several types and among the major ones are machine learning algorithms, computer vision algorithms, and natural language processing algorithms. The use of specific algorithms and models would depend on the kind of data they would need to learn from and interpret such as text, voice, and images. One must remember that the development of these building blocks would need creative problem-solving skills, an important programming skill that continues to be the domain of human coders.
Utilizing human intelligence is also much needed in other aspects of AI systems development. Depending on the area of application, having a human domain expert is very valuable as they possess in-depth knowledge of the specific field or industry which would come in handy in understanding the problem requirements, defining relevant features, and evaluating performance.
Human designers are able to put the human touch into the creation of user-friendly interfaces and the design of user interfaces. They are able to work closely with the development team to ensure that the AI system is intuitive, visually appealing, and meets user needs.
Humans are also better suited to better address ethical considerations in the development and deployment of artificial intelligence. These can include bias and fairness which can affect individuals and groups, privacy and security, and accountability and transparency which can lead to mistrust and misunderstanding.
So far, it is still obvious that human intelligence is essential for the development and deployment of AI systems. I share the belief that even if AI systems become more sophisticated, the need for human intelligence will only grow. If not, it will only be because humans have allowed AI systems to take control. If that happens, then, on their own, AI systems would be able to easily answer the question “Is that a white shirt with black stripes or a black shirt with white stripes?” ([email protected])
(The author is an executive member of the National Innovation Council, lead convenor of the Alliance for Technology Innovators for the Nation (ATIN), vice president of the Analytics Association of the Philippines, and vice president, UP System Information Technology Foundation.)