ChatGPT can now speak with emotions and even tell jokes — just like a human. While we might feel closer to artificial intelligence than ever before, chances are it's lying to us.
Category
🗞
NewsTranscript
00:00 ChatGPT just got more human-like. It speaks with emotion and perceives the
00:05 world through your smartphone camera. But at the same time, researchers warn that
00:11 AI still fabricates facts and can sometimes deceive humans. So will GPT 4.0
00:19 be your trusted helper or a convincing liar?
00:23 GPT 4.0 is OpenAI's newest model with "O" standing for "omni" as in "omnipotent".
00:31 An interesting claim. The model can work with text, audio and video, making the
00:36 interaction more natural. Let's take a look.
00:39 And I wrote one last thing I'd love if you could take a look at.
00:42 Of course, I'd love to see what you wrote. Show it to me whenever you're ready.
00:46 Okay, so this is what I wrote down. What do you see?
00:49 Oh, I see. I love ChatGPT. That's so sweet of you.
00:56 According to its makers, GPT 4.0 doesn't just sound emotional itself.
01:01 It can recognize your emotions, too.
01:04 It looks like you're feeling pretty happy and cheerful.
01:08 So what's it good for? It's meant to be your everyday assistant.
01:12 The chatbot could be used to translate instantly what you're saying into a different language.
01:18 In theory, enabling a real conversation.
01:21 Hey, how's it been going? Have you been up to anything interesting recently?
01:26 Hello, how are you? Have you done anything interesting lately?
01:33 However, AI models are known to not work that well in languages other than English.
01:38 So we'd have to see if that will change with GPT 4.0.
01:42 According to OpenAI, it can also help with tasks like preparing for job interviews,
01:48 giving information on your surroundings or judging your outfit with humor.
01:53 Oh, Rocky, that's quite a statement piece. I mean, you'll definitely stand out.
02:03 But what does it mean if AI adopts more and more human traits?
02:07 Can the model become indistinguishable from humans?
02:11 Trusting the algorithm.
02:14 We tend to trust a computer's decision, even if it contradicts our own beliefs.
02:19 That's what scientists found as early as 1999.
02:23 And more recent studies suggest the same.
02:26 Chatbots are already very convincing.
02:29 If they speak in a human-like voice, that might make it even easier to believe them.
02:34 That can be a problem because AI is lying.
02:39 Hallucination, or AI inventing facts it believes to be true, has long been discussed.
02:46 But AI can also be straight-up lying, even if it knows the right facts.
02:51 Multiple tests with AI playing board games, poker or video games showed
02:56 the systems have learned to bluff to produce a favorable outcome.
03:00 It's not unimaginable that AI could be used to trick us
03:04 into giving up valuable personal information like credit card details.
03:08 I have no doubt a lot of people would fall for chatbots' lies
03:12 if presented in a convincing human-like manner.
03:15 Especially if you factor in emotional bonds.
03:19 Recently a study showed that kids believe AI assistants like Alexa or Siri have feelings.
03:26 Some even said if their Alexa breaks down, it wouldn't be right to throw it away
03:31 and that it can feel left out in conversations.
03:34 Some people are already making friends with chatbots.
03:38 And the app Replica, for example, describes itself as "the AI companion who cares".
03:45 With the new tech, the emotional connection to bots will only get stronger.
03:49 And this could get ugly very quickly.
03:52 In 2021, a Replica user set out to murder the late Queen of England
03:57 after the chatbot repeatedly encouraged his delusions.
04:01 I for one am quite skeptical about feeding all my personal data to chatGPT.
04:06 I mean, if you use it like the demo suggests,
04:09 it will know things like your voice, face, house and schedule.
04:13 No thanks.
04:15 What about you? Would you use it as an everyday companion?
04:18 [ Silence ]