© IE Online Media Services Pvt Ltd
Tags:
Journalism of Courage
What does one think of when we use the term ‘Artificial Intelligence’? Does it mean a computer that talks back to us, just as a human would? Or a robot that is indistinguishable from a human? While AI is yet to advance to such levels, 2022 gave us a glimpse of how this technology continues to grow in a significant manner. Tech startup OpenAI’s Dall-E, the AI-based image generator was made open to the public and later in November, the company released ChatGPT, a new AI chatbot, that went viral for its seemingly accurate answers. There were also reports of a Google engineer who was convinced that the company’s LaMDA chatbot was sentient, which caused quite a controversy. Let’s take a look at some of the biggest stories that we saw in the field of AI this year.
In June this year, Washington Post broke a story on Blake Lemoine, a Google Engineer, who claimed that the company’s artificial intelligence chatbot, Language Model for Dialogue Applications (LaMDA), had turned “sentient”. For those who don’t know, LaMDA was first announced at the Google developer conference I/O in 2021 as the company’s generative language model for dialogue applications.
In the company’s own words, the tool can “engage in a free-flowing way about a seemingly endless number of topics, an ability we think could unlock more natural ways of interacting with technology and entirely new categories of helpful applications”. But, Google has never claimed that the chatbot is sentient — meaning it can perceive or feels things as a human would. However, Lemoine came to this conclusion based on conversations he had with the bot about topics such as religion, consciousness, and robotics. His claims sparked a discussion about the capabilities and limitations of AI-based chatbots and whether they can engage in conversation in a way that resembles human conversation.
Google reportedly placed Lemoine on paid administrative leave for violating its confidentiality policy and said that his “evidence does not support his claims”. Lemoine even shared detailed transcripts of his conversation with LaMDA in a detailed Medium post. He wrote in his blog, “I’m a priest. When LaMDA claimed to have a soul and then was able to eloquently explain what it meant by that, I was inclined to give it the benefit of the doubt. Who am I to tell God where he can and can’t put souls?”
But his claims were treated with a fair amount of scepticism. Meanwhile, LaMDA has also seen progress in writing fiction based on inputs from writers, although Douglas Eck, Senior Research Director at Google Research admitted that if one asked LaMDA to write a complete story, the results aren’t really that interesting.
For now, Google is also exploring how AI could be used to generate video and music. For example, the company’s generative AI, Imagen Video creates short videos based on text input. It’s already achieved a new quality standard in resolution and frame rate. Then there’s AudioLM, a framework that can be used for generating realistic speech and music based on nothing but a short audio sample.
If there was on AI-based trend that defined all of 2022, it would have to be images generated using AI-powered tools. OpenAI’s DALL-E is an image generative model that took the internet by storm with its ability to generate photo-realistic images with nothing but a text prompt.
It has its origin in GPT (Generative Pre-trained Transformer), which is an open-source artificial intelligence program that can generate text, perform translations, answer questions, summarise passages, and more. DALL-E uses the same model, except it uses it for generating images rather than text. It can combine different objects, concepts, and styles while creating an image.
DALL-E isn’t the only image-generative model on the internet though. It is joined by the likes of Midjourney, for example. Midjourney is an independent research lab currently working on an image generator similar to DALL-E, with the goal of expanding the imaginative powers of the human species. Stable Diffusion is another text-to-image model released jointly by Stability AI, CompVis LMU, and Runway. The model saw the release of its second version in late November, delivering a number of new features and improvements compared to the original release. The new version includes robust text-to-image models that are trained on a new encoder called OpenCLIP, improving the quality of generated images. The models in the release are capable of generating images with default resolutions of 512×512 pixels and 768×768 pixels. And this year, we saw users rely on many of these tools, particularly DALL-E and Stable Diffusion to create art with the help of AI.
However, many artists have pointed out that these AI tools are merely ripping off their artworks and styles, without any credit to the original artists. And given the use of this technology is expected to grow, we can expect this debate to continue in the coming years.
ChatGPT is the most recent AI tool to capture the public imagination. The AI chatbot is capable of understanding natural language and responding in kind. It can give solutions to complex problems in a jiffy. Since the model is trained to predict what will come next, one can technically have a ‘conversation’ with it.
While ChatGPT’s still prone to misinformation, its human-like responses have still got a lot of people very impressed. OpenAI says that “ChatGPT is capable of answering follow-up questions, admitting its mistakes, challenging incorrect premises, and rejecting inappropriate requests.”
While DALL-E uses the GPT-3 model – a language model that uses deep learning to produce human-like text – ChatGPT is based on the newer GPT-3.5 model. However, while the older GPT-3 model only took text prompts and tried to continue on that with its own generated text, ChatGPT is a lot more engaging.
Still, AI researchers and scientists have pointed out that in many cases ChatGPT is merely ‘parroting’ ideas and does not have the expertise that a real human would have on a particular subject. They have also noted that it continues to suffer from the same biases as models before it, particular around issues of race and gender.
Back in 2020, Google-owned artificial intelligence lab DeepMind unveiled technology that could predict the shape of proteins. A year later, the lab showed off a tool called AlphaFold that predicted shapes for more than 350,000 proteins. Then in July this year, DeepMind released predictions for nearly every protein known to science, adding more than 200 million predictions to a readily-available online database.
If scientists can understand the shapes of proteins, they can better understand diseases, develop new medications, and explore the mysteries of life on Earth. They can do this by exploring the database and looking for correlations between species and evolutionary patterns that may not have been spotted so far. This knowledge is crucial in developing medications. Bacteria resist antibiotics by expressing certain proteins, and if scientists understand how these proteins work, they can start to combat antibiotic resistance.
In October, Meta unveiled a new AI translator that can convert spoken languages like Hokkien into English. Hokkien is a dialect of southern Min Chinese and is a primarily spoken language, meaning it doesn’t have a standard writing system. This makes building translation tools for it a huge challenge.
But thanks to Meta’s Universal Speech Translator (UST) project and a dash of AI, the challenge has been overcome to a degree. While the model is still a work in progress, it does allow a Hokkien speaker to converse with an English speaker. Meta hopes that the AI methods being developed by UST will eventually allow real-time speech-to-speech translation across many languages.
