Imagine you go to a bookstore, and you notice and exciting cover. You pick the book, read the summary at the back, and the rave reviews. The plot seems intriguing enough, but when you check for the writer, it says “ by AI-something.” Would you buy the book, or would you think that was a waste of money? We will have those decisions moving into the future, and who will be responsible for such writings? But, that shows how AI is learning to play with words.
You may as well decide now if you will purchase content written by AI. That’s what the future will bring — AI is learning to play with words.
All of us have gotten used to chatbots and their limited capacity, but it appears their boundaries will be surpassed. Dario Amodei, OpenAI’s research director, informs us they have created a language modeling program which is very imaginative, to say the least. Its latest achievement was creating counterarguments and discussions with the researchers.
The program was fed a variety of articles, blogs, websites, and other content from the internet. Surprisingly, it managed to produce an essay worthy of any reputable writing service, and on a particularly challenging topic, by the way (Why Recycling Is Bad for the World).
Did the researchers do anything to help the program by providing specific, additional input? Certainly not. GPT-2, OpenAI’s new algorithm, did everything on its own. It excelled in different tests, such as storytelling, and predicting the next line in a sentence. Admittedly, it’s still far from inventing an utterly gripping story from beginning to the end as it tends to stray off topic — but it has great potential.
What sets GPT-2 apart from other similar AI programs is its versatility. Typically, such programs are skilled only for certain areas and can complete only specific tasks. However, this AI language model uses its input and successfully deals with a variety of topics.
What exactly can this AI program do?
For starters, the program could be used for summarizing articles or translating. Chatbots would be more informative and flexible. Ultimately, the program could turn into an excellent personal assistant, summarizing the reports for you, and sending out a proper company or business information. These types of programs would more or less be doing what a human PA already does.
There are several very negative possibilities.
At present, this type of AI program pretty much comes down to work best depending on what kind of material it is being fed with. For instance, if this AI program is given offensive, violent content. If the AI is given specific articles degrading individual nations or races (take some Hitler’s propaganda, for example), GPT-2 would respond similarly and produce vile and offensive content.
The system could be abused for automated and refined trolling.
People would be convinced they are receiving someone’s real opinions, instead of an automated response by a machine. Also, phishing schemes would become much easier to pull off. Certainly, the cyber hacks will be all over this AI code and work. The internet could be overfilled with artificial content. Do we want this as our only future? Let’s remember the example of pornographic deepfakes — a classic example of how top technology was abused when released into the public.
What is to come out of all the new AI code?
As the proponents of AI and AI code freely themselves admit — only the future will show what is really in the works at present. In the meantime, OpenAI will continue to invest resources in their innovative language model and feed it with more and more data, hoping for the best. Will we be on the receiving end of “the best?” The answer to that question will remain to be seen.