What is AI?

Artificial intelligence is the simulation of human intelligence by machines. Machines do not possess the same 'thinking' skills that we humans do. Instead, they process billions of calculations per second to determine the best way to reach a solution. Artificial intelligence analyzes large amounts of labeled data for patterns and correlations. It then uses the rules it has learned to extrapolate to other data. For example, if I feed a collection of 5000 images of different zebras to an AI, it will observe the pattern that all the zebras have black and white stripes. It will understand that pictures of zebras typically have half the pixels being dark and the other half white. When I give the AI an image of a zebra it has never seen before, it will be able to identify it as a zebra because it has learned the rules of what a zebra looks like.

Unfortunately, AI can still make mistakes. Even though machines have insane amounts of processing power, the rules that the system makes may not be entirely accurate. Returning to our zebra example, there are zebras in the world that are mostly black or mostly white. When the AI is fed an image of z zebra of this type, it uses the incorrect 50-50 pixel rule to determine that the picture is not a zebra. To account for these types of mistakes, scientists and researchers must provide a variety of images of different types to the AI. Larger amounts of labeled data lead to a higher accuracy.

The advantages and disadvantages of AI

Given that AI has immense processing power, it is helpful for routine, detail-heavy tasks that would otherwise take a long time to do. AI is used to detect certain cancers such as breast cancer and melanoma. The relatively low mistake rate makes AI more efficient than doctors. In the field of finance, AI is used to increase the speed of processing transactions. Using information such as purchase time, location, and item name, AI can effectively detect fraud.

Doctors use AI to monitor patients' vital signs and alert medical professionals when discrepancies are detected. Removing the need for a nurse to be near the patient at all times, AI has increased hospitals' ability to admit more patients and take closer care of more people. This was especially useful during the COVID-19 pandemic. Healthcare systems use AI to recommend treatments for patients. Using information from the patients about their symptoms, AI can determine the patient's infection and suggest the best medicine in real-time. During emergency situations, this service provides immediate care when medical professionals are unavailable. Doctors also use AI to create new drugs. After analyzing past combinations of useful drugs on the market, AI can find promising new drug designs and combinations.

AI is incredibly useful, but it still has some disadvantages. AI is expensive to develop and maintain. It is only as useful as the data it is trained from, and the data it is given must be first analyzed by humans. AI also cannot 'think outside the box' and can only achieve specific results without being reprogrammed or retrained. It retains biases from the training set and can still make mistakes if the data is not diverse enough. AI also has the potential to replace some jobs, resulting in higher unemployment.

So how does ChatGPT work?

Unlike other types of AI, ChatGPT interprets and responds to any question asked. Just like other types of AI, ChatGPT uses patterns to provide answers. ChatGPT has been fed enormous quantities of labeled text data from billions of webpages online. It has then computed relationships between the words and found patterns in language. Basically, ChatGPT is predicting the next word to output based on the last word it has chosen. Due to this mechanism, ChatGPT may not have correct facts - the next word outputted is the best possible prediction even though it is not true. After all, AI is not self-aware and cannot understand what it is 'saying'.

In Microsoft Bing's spinoff of ChatGPT, Microsoft has allowed the AI to access current website articles. By training the AI on recently created websites, the AI is now aware of current events and newly discovered information. While ChatGPT is unable to speak on more recent topics, Bing's AI can provide this information with cited sources, making it more useful to the general public.