Creativity or delusion? Don’t panic when facing AI hallucinations

Recently, Xiao Ai captured the guessed answer by searching the search engine information and generated the answer, which attracted a lot of attention. Artificial intelligence intentionally or unintentionally fabricates, which is called AI illusion in the industry. As a marketing planner, how should we deal with this situation?

In this era where revelations are often reversed, it is not uncommon to debunk rumors, but the reversal of Xiaomi's price incident still attracted a lot of attention. Recently, a screenshot of the price of Xiaomi's car models circulated on the Internet. The leaked picture shows that the Xiaomi model is priced from 149,900 yuan and has a maximum range of 800 kilometers. However, not long after, Xiaomi's public relations director posted a Weibo post to refute the rumor.

It turns out that this is the answer generated by Xiao Ai by retrieving search engine information. In fact, this information is conjecture information reported by self-media in May 2022. When someone asked Xiao Ai a question, Xiao Ai simply captured this conjecture information.

As Xiaomi's artificial intelligence assistant, it is not surprising that the answers given by Xiao Ai have attracted attention. Indeed, Xiao Ai cannot distinguish the authenticity of information, which is unintentional. However, this incident reminds me of the phenomenon of artificial intelligence deliberately making up stories, which is called AI illusion in the industry.

Can AI also have hallucinations? The era of artificial intelligence has triggered a carnival of productivity liberation, but behind this, the crisis of over-reliance on artificial intelligence seems to have emerged!

1. What is AI Hallucination?

Have you ever encountered an AI tool that said nonsense in a serious manner?

In the field of artificial intelligence, AI "hallucinations" are usually defined as "generated content that is meaningless or unreliable compared to the source content provided." That is, when a model has "hallucinations," the content and data it outputs may be deceptive, and the training data it uses cannot prove the rationality of the output results.

Simply put, AI sometimes generates fictitious people or things and maintains consistent confidence. This false information is highly misleading and deceptive.

Even the CEO of OpenAI publicly stated on Twitter that "ChatGPT does know a lot of things, but the dangerous thing is that it is confidently wrong a considerable part of the time."

However, the word "illusion" always gives people a hint that AI has vitality and thinking ability. Because of this, as more and more people pay attention to this phenomenon, this label is becoming more and more controversial. I think that AI "illusion" is actually a kind of information fiction, which is based on the AI ​​model, not the subjective thinking ability or creativity of AI.

Regardless of what it is called, the phenomenon of self-deception and fabrication in artificial intelligence is no longer an isolated case. Under the wave of AIGC, the use of AI tools seems to have become the new normal in the new era.

2. Diseases that AI can easily reveal

This phenomenon of AI making mistakes or deviations, leading to inaccurate, erroneous or even absurd results, can occur at multiple levels, such as the fabrication of text, the false fabrication of facts, the generation of useless code, or even giving you a fabricated reference.

The Washington Post reported that a law professor discovered that ChatGPT included him in a list of legal scholars who had sexually harassed someone. There are also reports that ChatGPT fabricated an Australian mayor who was found guilty of bribery and sentenced to prison! Isn’t it terrible? With the rise of AIGC, if this AI illusion is not effectively resolved, we may be swept up in all kinds of fabricated information.

3. Why does AI produce this illusion?

As I mentioned above, understanding the nature of large language models (LLMs) can help us find the answer. The design of large language models is based on the "statistical probability" of language, which means that AI has no "real world experience" at all and cannot make judgments based on the truth or falsehood of facts. Therefore, AI does not know what is correct and what is incorrect, and can only maintain statistical consistency with the prompt words.

"Even the most advanced AI models are prone to lying, often showing a tendency to fabricate facts in moments of uncertainty," OpenAI researchers wrote in the report. "These hallucinations are particularly severe in domains that require multi-step reasoning, where a single logical error can be enough to undermine a larger solution."

However, from another perspective, this illusion is precisely a manifestation of AI's creativity. Because it is this ability to respond creatively that makes AI a powerful tool for generating new ideas or unblocking bottlenecks in thinking. It is useful when you use it for brainstorming and inspiring inspiration, but you need to be cautious when using it as a factual reference.

As marketers, we should remain cautiously optimistic when using such tools, because a lot of the work we do is creative or innovative, and AI illusions can bring more inspiration and enlightenment, but there is still a lot of work based on objective facts, which requires cross-verification or verification, rather than leaving it all to chance.

It can be seen that AI tools have indeed liberated productivity in terms of efficiency, but in terms of content authenticity, we still need to have basic recognition and judgment capabilities when using them. After all, tools are always used by people, not to completely replace people. Do you feel that your career is stable?

I hope the above can give you inspiration.

Author: Little Monk Kunkun, WeChat Official Account: Marketing Zen Meditation Center