Week of January 22, 2023
Artificial Intelligence: An almost surreal article appeared in the news last week: “Member of Congress reads AI-generated speech on House floor”. Jake Auchincloss (D-MA) decided to use a non-standard method of delivering a speech on a bill that would create a U.S.-Israel artificial intelligence center: He made use of the now-famous Chat-GPT3 artificial intelligence software. Auchincloss said that he prompted the system to “write 100 words to deliver on the floor of the House of Representatives” about the legislation. While it required a few iterations, he eventually came up with something that could be delivered on the floor.
GPT-3 (Generative Pre-trained Transformer 3) is a state-of-the-art language processing Artificial Intelligence (AI) model. GPT-3 processes text input to perform a variety of natural language tasks. It uses both natural language generation and natural language processing to understand and generate natural human language text. GPT-3 can create anything with a text structure – not only human language text. It can also generate text summarizations and even programming code. It is probably the most comprehensive and most powerful language processing AI model ever created. GPT-3can imitate human speech in a way that makes it nearly impossible to know that the speech is being generated by a machine. It passes the Turing Test – the test of a machine's ability to exhibit intelligent behavior equivalent to, or indistinguishable from, that of a human being – with flying colors.
AI has come a long way since its inception. The AI that powers GPT-3 is based on “deep learning,” in which a “neural network” is trained to recognize and recreate patterns. Deep-learning-based AI has exploded in the last generation, due to drastic increases in computing power and in the sheer amount of data available to train the network. For example, to teach an AI algorithm to recognize a bear, it needs to be shown pictures of bears. The more pictures it sees, the more different kinds of bears it sees, and the more angles it sees them from, the easier it is for the algorithm to recognize bears. The internet is a treasure trove of bear pictures. After seeing millions and millions of images of bears, AI algorithms can pick them out where most humans cannot. For example, the famous Tobelerone chocolate bar, made in Bern (German for “bear”), Switzerland, has a bear hidden in the logo. I’ve seen Tobelerone bars for years and I never saw the bear.
An AI algorithm picked it out immediately. Can you?
In 1997, IBM’s Deep Blue AI program beat chess grandmaster Garry Kasparov, arguably the greatest player ever. All that was required was feeding the algorithm that powers Deep Blue all of the moves in a huge number of chess games, games that were won and lost. The computer slowly began to “understand” which moves led to victory and which moves led to defeat. Each time it played, whether it lost or won, it learned from its experience. Now, no grand master can hold his own against a good AI-based chess program.
AI shines in nearly anything that requires pattern matching, such as locating a hairline fracture in an x-ray or a tumor in a CT scan. Most experts agree that in a few years, many tasks that are currently performed by trained professionals, such as radiology, will soon be performed better, more quickly, and more cheaply by computers. GPT-3, like all AI algorithms, was trained using large amounts of data to match patterns, predicting which words, phrases, and topics are “expected” in a sentence or in an article. GPT-3 was actually fed the entire internet in order to learn how to speak.
Let’s take a closer look at the substance of Auchincloss’s speech. “You are Jake Auchincloss, a Member of Congress. Write 100 words to deliver on the floor of the House of Representatives. Topic: the importance of the United States–Israel Artificial Intelligence Center Act, which the congressman will re-introduce this term.” Here is what GPT-3 eventually wrote: “Madam Speaker, I stand here today because I am planning to re-introduce the United States-Israel Artificial Intelligence Center Act, a bipartisan piece of legislation that will cement a mutually beneficial partnership between the United States and Israel on artificial intelligence research. This is a critical step forward in an era where artificial intelligence and its implications are taking center stage in public discourse. We must collaborate with international partners like the Israeli government to ensure that the United States maintains a leadership role in AI research and development and responsibly explores the many possibilities evolving technologies provide. The United States-Israel Artificial Intelligence Center Act will allow us to tap into the expertise of both countries and draw upon each other’s resources to explore and develop cutting-edge AI advancements.”
The “United States–Israel Artificial Intelligence Center Act” was introduced as bill S.2120 by a bi-partisan group of Senators: Marco Rubio (R-FL), Maria Cantwell (D-WA), Jacky Rosen (D-NV) and Marsha Blackburn (R-TN). The bill sought to “require the Department of State to establish the United States–Israel Artificial Intelligence Center to leverage the experience, knowledge, and expertise of institutions of higher education and private sector entities in the United States and Israel to develop more robust research and development cooperation in specified areas (e.g., machine learning, object detection, and speech recognition).” While the bill had traction in the Senate, its sister bill in the House, H.R. 5148, did not. Congressman Auchincloss is trying to reintroduce the bill, to put the wind back into the sails of joint U.S – Israel AI Collaborative Research. USIEA promotes education on Capitol Hill regarding the issues surrounding AI and sees it as a jumping-off point for a new and exciting lane of action.
Wishing you a quiet week,
Ari Sacher
