Google’s executives discussed the growth of ChatGPT in an all-hands discussion however, they noted that the outstanding issues with AI chatbots such as bias and accuracy indicate they’re not fully ready to replace the search engine.
ChatGPT’s launch ChatGPT has caused people to think that AI chatbots will be able to take over conventional search engines. However, executives at Google claim that it’s still unproven to be put in front of people, and has issues like chatbots’ biases toxicity, and a tendency to simply create data.
According to an article from CNBC, Alphabet CEO Sundar Pichai and Google’s chief of AI Jeff Dean addressed the growth of ChatGPT at an all-hands discussion. One employee asked if ChatGPT’s launch -created by OpenAI which has close connections to Google competitor Microsoft was the possibility of a “missed chance” for the giant search engine. Pichai and Dean were reported to have replied that Google’s AI language models are as powerful as OpenAI’s however, they said that the company needed to be “more prudently than a small start-up” due to its “reputational risk” that the technology could pose.
“We are determined for ways to incorporate these capabilities into products that actually work and things that are prominently incorporating the model of language rather than hidden under the covers as we’ve used them up until now,” said Dean. “But it’s crucial that to get this right.” Pichai added that Google has “a large amount” in the works for AI features that will be available in 2023. He also said it is “this will be an area in which we must be courageous and responsible, and we must balance the two.”
Google has created a variety of massive AI model languages (LLMs) comparable in terms of capability similar to the OpenAI ChatGPT. They comprise BERT, MUM, and LaMDA and all have been utilized to improve google’s Search Engine. These improvements aren’t obvious, but they are focused on analyzing user queries to better understand what they are looking for. Google states that MUM assists in determining the signs that the user is experiencing an emotional crisis for instance and then directs them to helplines or information from Samaritans groups. Google has also introduced apps such as AI Test Kitchen in order to offer users an opportunity to experience its AI chatbot technology, however it has hampered interaction with users in a variety of ways.
OpenAI is, too, initially cautious when it came to developing it’s LLM technique, yet it has changed strategy with the release of ChatGPT and made it available to everyone. It has resulted in an explosion of positive publicity and excitement for OpenAI even though the company incurs massive costs to keep the system available for free.
While LLMs such as ChatGPT exhibit remarkable flexibility in creating language, they are not without their flaws. The way they do it is to amplify social prejudices that are found in their data for training which often slam women and other people who are of color. They can also be very easy to fool (users discovered that they were able to bypass ChatGPT’s safety rules which are intended to prevent it from providing risky information, through asking the system to assume it’s a bad AI) Also, andprobably most relevant for Google the company — they frequently provide misleading and false responses to questions. Users have discovered the fact that ChatGPT “lies” on many different questions, from faking biographical and historical information, to justifying dangerous and false assertions, such as saying that the addition of crushed porcelain to milk “can help the infant’s digesting system.”
In the Google all-hands meeting Dean acknowledged the many issues. He stated that “you could imagine that for search-related applications that the issues of factual accuracy are crucial, but in other cases, bias as well as toxicities and safety concerns are just as important.” Dean declared it was true that AI chatbots “can create a false impression. […] When they’re unsure regarding something. They’ll say, for instance elephants are the ones who lay the most eggs, or whatever.”
Although the announcement of ChatGPT has brought about new debates about the potential for chatbots to take over the traditional engines for searching, this question was being debated at Google for quite some time, and sometimes it has caused controversy. AI researchers Timnit Gebru and Margaret Mitchell were dismissed from Google after they published an article that outlined the ethical and technical challenges that come with LLMs (the same issues which Pichai and Dean have been describing to the staff). In May of the year before, a team of Google researchers delved into the same issue about AI in search and identified numerous issues. According to their research paper one of the most significant problems has to do with the fact it is that LLMs “do not possess an accurate comprehension of the world around them and are susceptible to dreaming, and most importantly, they cannot justify their statements by referring to supporting documents from the corpus they learned about.”
There are methods to address these issues, of course and other tech firms will surely be considering whether the possibility of launching an AI-powered search engine, even risky one is worth the effort to get a leg up on Google. In the end, if you’re just starting out, “reputational damage” isn’t as important.
For the part of OpenAI it appears to be trying to dampen expectations. The the CEO Sam Altman recently tweeted: “ChatGPT is extremely limited, but it’s good enough at certain things to give an impression of its greatness. It’s a mistake dependent on it for anything of importance at the moment. It’s a sign of progress but we still have plenty of work to be done on reliability and honesty.”
