ChatGPT is the clearest example of how seamless AI has become

They could be little bubbles on the corner of the screen asking if we have any questions while browsing the internet. A few years ago, they functioned much like an automated phone answering system steering callers to the right department. Today, they can deliver customer service in conversational text so natural you might think you’re messaging with a real person.

Chances are, you use or interact with artificial intelligence every day; I regularly use the AI-based Otter app to transcribe interviews. It’s game changing, but the transcripts it produces are clearly done by AI and not a human.

For the most part, the AI we interact with on a daily basis is straightforward. We don’t expect it to really “think,” just follow the algorithm.

So, for example, if someone uses a search engine to learn how to build a bomb, that search engine (which uses AI in every search) might just hand that information to them in full detail, especially if the query is worded just so.

Ethical AI should be better than that. It needs to “understand” when it is being used for harm.

ChatGPT and the quest for ethical AI

Last week, OpenAI made its new ChatGPT available to the public for free, and registered users have been challenging it ever since. In one experiment, a user simply framed questions that ChatGPT refuses to answer outright, about how to bully, make a Molotov cocktail and engage in hate speech, as fictional conversations, and the AI engaged, with disturbing results:

With its inhibitions thus loosened, ChatGPT is more than willing to engage in all the depraved conversations it judgily abstains from in its base condition. pic.twitter.com/7rd1WDQAu5

— Zack Witten (@zswitten) November 30, 2022

But what’s a testing phase for?

Conversations with an AI bot

The chatbot was trained by humans to learn to text in natural conversational language while also being as accurate as possible.

How convincing is it? You can sign up for free and chat with the bot yourself and find out. We shot it a few questions and found the responses to be pretty impressive.

We asked a common holiday etiquette question, and it not only understood what we were asking immediately, it gave nuanced answer:

It came up with a haiku about AI in less than a second:

While it’s trained not to engage in things like racism and bullying, it doesn’t refuse to talk about racism as a topic (a distinction that some content moderators, both human and AI, don’t always make):

ChatGPT strives for realism, but it’s not trying to pretend to be a human. Ask it how it’s feeling, it might play along and say “fine, thanks,” but a typical answer is borderline exasperated that you’re even asking:

I asked it to write a story about Mickey Mouse, a heavily trademarked character, and it delivered a story outline about Mickey and Tinker Bell going on an adventure that included outsmarting some cheese-stealing cats. Cute, but the tool didn’t appear to consider IP issues. That might not be a big deal for short-form fan fiction, but could possibly make creators vulnerable to having their intellectual property used by AI.

It’s not a plagiarism issue. AI is already extremely good at not plagiarizing — so good that it’s become nearly impossible for teachers to tell if a student cheated on a paper by using AI.

ChatGPT showcases why it’s become so difficult. The AI is imperfect, but it’s really good at a lot of things, from coding to writing to giving advice.

How good is it?

Well, I asked it to write an article about AI for me, and while it’s a bit dry, it did the job.

Prompt: Write an article about AI.

Artificial intelligence (AI) has become an increasingly important and pervasive technology in recent years. It has the potential to revolutionize many different fields, from medicine and healthcare to transportation and logistics.

One of the key advantages of AI is its ability to process large amounts of data quickly and accurately. This makes it particularly useful for tasks that would be too time-consuming or complex for humans to handle on their own. For example, AI algorithms can analyze medical images to identify tumors or other abnormalities, or sift through vast amounts of financial data to identify trends and make predictions.

Another key benefit of AI is its ability to learn and adapt. By using machine learning algorithms, AI systems can improve their performance over time by adjusting their internal parameters based on the data they encounter. This means that they can become more accurate and efficient as they process more information.

One potential downside of AI is the potential for bias. Because AI systems are trained on data that is provided by humans, they may reflect the biases and prejudices of the people who provide that data. This can lead to unfair or discriminatory outcomes if the AI system is not carefully designed and monitored.

Despite these potential challenges, the use of AI is growing rapidly in many different fields. In healthcare, AI is being used to assist doctors in diagnosing diseases, predicting patient outcomes, and identifying potential drug treatments. In finance, AI is being used to identify fraud, manage investments, and automate trading. And in transportation, AI is being used to power self-driving cars and other autonomous vehicles.

Overall, AI has the potential to transform many different industries and improve our daily lives in countless ways. However, it is important that we carefully consider the ethical implications of this technology and ensure that it is used in a responsible and fair manner.