Artificial intelligence may have started with Alan Turing's speculations on intelligent machines in 1950. Natural language processing and machine learning have revolutionized industries that rely on big data, such as translation tools, financial institutions and search engines. However, there’s still a considerable gap between the complex algorithms that predict behavior based on massive amounts of data and the representations of AI in science fiction, which include Star Trek’s erudite Lt. Cmdr. Data and the rogue computer HAL 9000 in the move 2001.
Despite continuing advances in the field, the technology has not yet achieved the level of complexity needed to mimic the human mind. Chances are, AI will always need human inputs to govern the ethical considerations of leaving decisions up to the bits and bytes inside a computer algorithm.
Let's take a look at three reasons why human input guides AI use in established applications and emerging technology. As of today, and for the foreseeable future, even the most powerful AI requires human intervention to create meaningful outcomes.
1. Introducing Creativity and Compassion
When Apple first release the iPhone, no one predicted the number of jobs this one innovation would create. Whole industries sprang up around creating smart devices based on Apple or Android platforms. E-Commerce websites, mobile applications, wearable technology, online communities, and even ride-sharing emerged. Apple’s technology helped create the backbone upon which the tech industry grew. However, it took creativity for people to find useful ways to expand the technology.
The world of AI is desperately in need of the same innovation to create useful applications and improve the predictive capability of AI and deep learning. People will drive the creativity and compassion needed to use AI to create music, art and poetry, for example.
Kai Fu Lee stated in a Ted Talk that humans who wish to thrive in the coming AI age should harness creativity and compassion and use it to improve the performance of thinking machines.
2. Providing Governance
Inside and outside the digital world, planning consists of outputs, inputs and processes. Without this structure, robots can do little on their own. For example, when programmers change a web form, it throws off automatic processing, which hasn't been updated to accommodate new information or logic resulting from the change.
To prevent the same kind of thing from happening with AI and machine learning, we need human intervention to map out how machines work together. When employees come to work and clock in for their shift, managers can see this information by looking it up in the HR or timekeeping application. However, it's much harder to tell whether automation robots are on the job. For example, if someone changes a system password, the bot may not be able to complete the authentication process. Therefore, they aren't able to complete their assigned task and no one may notice for hours, days or weeks. Tasks performed by AI need to be monitored as closely as those performed by human employees. This governance requires a human touch.
3. Data Manipulation
You would think that data manipulation is a perfect task to hand over to an army of AI robots. However, that couldn't be further from the truth. Data collection, annotation and validation by people helps improve the performance of AI algorithms.
Machine learning algorithms handle sound, video and text, each of which requires unique handling. Curating high-quality data continues to be a central challenge to the widespread use of AI for task automation. Many algorithms use transfer learning techniques. The technique reduces the need for continuous discrete data, which helps the algorithm improve its predictions. However, deciding which data to include requires human intelligence and selection processes.
Human data collectors ensure that the AI technology has a consistent flow of divergent data points to help improve performance. In translation applications, voice recognition tools must learn new industry jargon or slang words in order to remain effective for the widest possible set of users. Otherwise, more and more words will confuse the algorithm, resulting in a garbled translation.
A system can collect millions or billions of unorganized data points, but they aren't very useful without someone to train the AI. Annotating data points helps the algorithm correct false assumptions and build its vocabulary. Annotations include categorization, semantics and other information that can describe text, videos and images.
Test users chosen to evaluate AI systems should be from the target audience or as close as possible to them. This helps ensure that the end-user will get consistent results from the AI. In translation tools, this would include linguists who speak the language and native speakers who can catch the nuances required for effective communication.
As seen in the examples above, AI will always need some kind of human touch. AI supersedes human capability when it comes to crunching enormous amounts of data for calculations and trends. However, we aren't at a point where humanity should worry about thinking machines becoming independent of their creators.
AI will continue to rely on human intelligence and creativity to direct and correct its assumptions. AI can predict our preferences online and forecast the probability that we will break traffic laws. However, it doesn't yet have the unique perspective needed to replace our intuition and common sense.