Hop onto the “AI wave” (Part 1)
The year in which AI awes took the internet by storm: 2022 (A year in review)
As we crossed the mark of 8 billion individuals in the preceding year, we established our potential to exist as a species. But it’s not just the number of inhabitants, more astonishing certitude is our aptness to mimic our own intelligence as technology. Whether you follow some tech news or not, We bet you’ve heard or overheard debates about AI breaking the norms over the internet.
In 2022 we have seen chatbots doing homework, answering complex philosophical questions and apps generating creative art. While the speculations of groundbreaking implementation of AI have been squirming among the enthusiastic community for a long time, recent developments have demonstrated that AI winter has met its closure.
“As the winter holiday approaches, it occurs to me that, instead of facing AI winter, we are in a boiling-hot summer of AI.” ~ Andrew NG (wrote as his opening lines, in his weekly issue of Dec. 21 2022, published in The Batch )
The recent advances might seem sudden, but “Rome was not built in a day”. They are the products of years of hard work and persistent research. Generative AI(software generating human-levelled text and images) tops the list of breakthroughs in 2022. While the social impacts, contextual understanding and ethical issues are still the areas of discourse, let’s look at some notable affairs that happened in the prior year from the lenses of an unprejudiced.
The one when an AI-generated picture won an art prize:
AI-generated art has been around for years but for the first time in history, an art submitted by Jason Allen(created using Midjourney) at Colorado State Fair took the blue ribbon home in the “digital art” category. As soon as this news landed on social media in August, it sparked backlashes and controversial debates around the globe.
Advancements in text-to-image generating apps like Midjourney, OpenAI’s DALL.E 2 and Stable Diffusion have surely changed the game of creative art. These apps based on diffusion models achieved an unprecedented degree of photorealism this year.
The one when everyone went crazy about “Magic Avatar”:
In November, Lensa.ai which uses Stable Diffusion under the hood to generate “magic avatars” from selfies, secured first place on iOS App Store in its category as the social media platforms got flooded by its AI-generated avatars.
The one when AI had feelings:
On June 11 2022, contemporaneous Google Engineer Blake Lemoine went public sharing his conversation with Google’s Language Model for Dialogue Applications (LaMDA) claiming it to be sentient. This claim provoked several Twitter debates. Later, Google’s spokesperson denied the claims saying “Of course, some in the broader AI community are considering the long-term possibility of sentient or general AI, but it doesn’t make sense to do so by anthropomorphizing today’s conversational models, which are not sentient. These systems imitate the types of exchanges found in millions of sentences, and can riff on any fantastical topic”. In simple terms: there is so much data, AI doesn’t need to be sentient to feel real.
The one when Meta had to take a step back:
In the middle of November, Meta released a demo of Galactica, a large language model trained on 48 million scientific articles. Not only did Galactica fail to address the problems of bias we have seen in other language models, but it also specialised in producing authoritative-sounding scientific nonsense. Two days later, amid controversy regarding the model’s potential to generate false or misleading articles, the company withdrew it.
The one when home-work is supposed to be dead, “ The ChatGPT Mania”:
One week after the withdrawal of Galactica, OpenAI demonstrated its marketing prowess and launched a public demo of ChatGPT (the latest in the research lab’s line of large language models of a class referred to as GPT-3.5). Soon after hitting the ground netizens went crazy and the screenshots of ChatGPT’s functional dexterities went viral on social media platforms and the mania is still ongoing.
The one when we knew all the protein structures:
On July 28 2022, DeepMind, Alphabet’s artificial intelligence subsidiary, announced that they have expanded their Alphafold database from nearly 1 million structures to over 200 million structures — visualizing nearly all catalogued proteins known to science. AlphaFold is trained on data in the Protein Data Bank, a global database of protein information. Visualizing Proteins’ shape and structure would replace the cumbersome process of laborious X-rays and microscopic examination turbocharging drug discovery.
The one when programmers found their best friend:
LLMs fine-tuned on computer code proved capable of generating software routines similar to the work of experienced developers.
In February, DeepMind introduced AlphaCode, a transformer pretrained on 86 million programs in 12 programming languages and fine-tuned on entries to coding contests. AlphaCode solves problems by generating millions of diverse programs using specially trained transformer-based networks and then filtering and clustering those programs to a maximum of just 10 submissions. AlphaCode claimed to achieve an average ranking in the top 54.3% in simulated evaluations on some programming competitions on the Codeforces platform.
In June, GitHub opened access to Copilot, an autocomplete system built on OpenAI, that suggests code in real-time. The tool works similarly to other AI software, such as ChatGPT and Dall-E 2, wherein it analyzes large troves of data, much of it culled publicly from the internet. But experts said Copilot is most notable because it’s at the centre of a lawsuit(claimed to be “one of the most important lawsuits happening right now”) that essentially calls that kind of learning a form of piracy. Matthew Butterick, a programmer and lawyer(part of a team filing a class-action lawsuit against Microsoft and other companies behind the tool) claim millions of programmers who wrote the original code, Copilot is trained on are having their legal rights violated.
What lies ahead?
“I probably started this year expecting 2022 to be more of the same, but there’s been some fairly ground-shaking developments this last year which I wasn’t expecting.” ~Peter Clark, the interim chief executive of the Allen Center for AI
While the recent implementations of AI in 2022 have certainly proven to leap over the “AI hype”, it is still just a few drops in the ocean. The technology is still a toddler as a lot of dystopian concerns like biases, ethical maturity, plagiarism, job losses and other psychological as well as societal implications are yet to be addressed. AI is going to change the basic paradigms of almost every industry and we have a lot to catch up in 2023 which we will talk about in the second part.
Bonus:
To demonstrate and to gauge the evolution of AI in last year, I tried an AI-based content generator Copy.ai to write this article. It is very simple to use and can also be used to generate captions for social media. Here’s the link to a notion page containing a fast-paced screen recording and the AI-generated version of this article:
References:
- https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai-in-2022-and-a-half-decade-in-review
- https://www.washingtonpost.com/technology/2022/12/28/ai-chatgpt-dalle-year-in-review/
- https://www.forbes.com/sites/bernardmarr/2021/09/24/the-7-biggest-artificial-intelligence-ai-trends-in-2022/?sh=202737332015
- https://www.deeplearning.ai/the-batch/issue-174/
- https://www.deeplearning.ai/the-batch/issue-176/
- https://www.deeplearning.ai/the-batch/issue-172/
- https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html
- https://www.washingtonpost.com/technology/2022/06/11/google-ai-lamda-blake-lemoine/