Five recent breakthroughs in AI you probably missed

The development of artificial intelligence has been extremely successful over the past two years as engineers and researchers demonstrate how AI can be used to enhance nearly anything.

According to Stanford's 2022 AI Index Report, private equity investments in artificial intelligence reached $93.5 billion in 2021. The International Data Corporation predicts that by 2024, investments in AI would total $500 billion. By 2030, according to PwC's forecast, the global AI sector would earn $15.7 trillion.

Artificial intelligence is swiftly establishing itself in new domains, including health, robotics, agriculture, and many more, thanks to years of significant investment by both public and commercial interests.

Even while learning machines are always getting better, AI has not yet lived up to the lofty expectations that most people have for it. While an AI can be exceptionally good at a narrowly focused activity, artificial general intelligence (AGI) is still now unattainable.

For every AI researcher, creating an artificial intelligence (AI) that can comprehend, learn, or perform any work that humans can is the holy grail. Algorithms, however, fall short in the actual world when complexity and the possibility of change are present.

Despite these drawbacks, considerable advances in artificial intelligence are expected in a number of disciplines by 2022 thanks to the billions of dollars that corporations like Google, Facebook, and Amazon have invested in the subject.

 

1. AI is amused by jokes

After Google revealed a 540 billion parameter AI natural language processing model that performs better than average, artificial intelligence made its single biggest advancement. Larger models perform better at 'transfer learning,' which aims to train neural networks with less data and computing power.

A variety of jokes PaLM from Google was able to interpret and explain. (Google) With 175 billion machine learning parameters, this replaces competitor OpenAI's GPT-3 natural language processing model. GPT-3, which was first introduced in May 2020, was praised for being so precise that it was difficult to tell whether the text was written by a human. Google's PaLM claims to be able to explain why a joke is funny, and it is now focusing on difficult problems of common sense and reasoning.

2. AI becomes artistic

DALLE 2 is a brand-new AI system created by OpenAI that can create art and realistic visuals from a text description you supply. The system's second edition, which was first unveiled in 2021, produces imaginative visuals with a resolution that is four times higher.

DALLE 2, which was made to show how AI perceives the environment and encourage creative expression, immediately prompts the query of what it means to be human and creative. This advances prior developments in AI-assisted visual processing technologies, such as the ability to colorize monochrome images or create lifelike 3D representations of people from ancient photographs. The most recent "multimodal" system, DALLE 2, is capable of working with both text and graphics.

3. AI may reduce the cost of streaming

The firm behind MuZero, which was able to master challenging games like Chess, Shogi, Atari, and even strategy games like Starcraft, is Deepmind. Deepmind is the same company that brought you AlphaGo, which defeated Lee Sedol, the current world champion in the game of Go.

The US Air Force would deploy Artu, which was created by the same corporation, as a radar operator, co-pilot, and mission planner for espionage planes. Recently, MuZero took on the problem of video compression and reduced the amount of data needed to transmit a video by an average of 4%. Given that typical compression codecs were developed after many years of work, that's no insignificant number.

According to analysts, the majority of internet traffic in 2021 was from video streaming. More efficient video compression could lower streaming prices and boost download energy efficiency in the years to come, when video streaming is only anticipated to increase.

4. The AI self-driving race intensifies

You can watch this year's competition between Tesla, Waymo, and Cruz self-driving cars as they travel from San Francisco to New York. Tesla plans to replace radar sensors with vision-based technology in a battle between competing approaches.

Radar, as opposed to cameras, can see through snow and fog, indicating a high level of confidence in Tesla's AI. More sensors have been installed by rival companies, and both strategies will now be compared.

LiDAR is a technology used by Waymo that is comparable to radar but uses laser pulses rather than radio waves. However, Tesla asserts that sensor data is simpler to interpret and provides a multidimensional image of the road that outperforms human reaction and visibility. The business just declared the results of its successful autonomous taxi trial in San Francisco.

5. Sensible optimism

Finding a technique to combine knowledge from various sources into reliable outputs remains a difficulty for most AI researchers, despite the fact that hybrid systems are slowly replacing specialization in AI systems.

Large-scale expenditures by huge tech companies in voice assistants like Alexa and Siri are accelerating the development of language models, which are used to communicate with people, respond to questions, and write convincingly.

While the year 2022 is far from over, there is a strong tendency toward merging speech, perception, and language difficulties rather than recognizing them as separate abilities. That's more difficult than it seems, but if recent years are any indication, artificial intelligence will continue to spur innovation for many years to come.