CodeXThe Stanford Center for Legal Informatics and the legal technology company Casetext recently announced what they called a watershed moment. Research collaborators had deployed GPT-4, the latest generation Large Language Model (LLM), to takeand passthe Uniform Bar Exam (UBE). GPT-4 didnt just squeak by. It passed the multiple-choice portion of the exam and both components of the written portion, exceeding not only all prior LLMs scores, but also the average score of real-life bar exam takers, scoring in the 90th percentile.

Casetexts Chief Innovation Officer and co-founder Pablo Arredondo, JD 05, who is a Codex fellow, collaborated with CodeX-affiliated faculty Daniel Katz and Michael Bommarito to study GPT-4s performance on the UBE. In earlier work, Katz and Bommarito found that a LLM released in late 2022 was unable to pass the multiple-choice portion of the UBE. Their recently published paper, GPT-4 Passes the Bar Exam quickly caught the national attention. Even The Late Show with Steven Colbert had a bit of comedic fun with the notion of robo-lawyers running late-night TV ads looking for slip-and-fall clients.

However for Arredondo and his collaborators, this is serious business. While GPT-4 alone isnt sufficient for professional use by lawyers, he says, it is the first large language model smart enough to power professional-grade AI products.

Here Arredondo discusses what this breakthrough in AI means for the legal profession and for the evolution of products like the ones Casetext is developing.

What technological strides account for the huge leap forward from GPT-3 to GPT-4 with regard to its ability to interpret text and its facility with the bar exam?

If you take a broad view, the technological strides behind this new generation of AI began 80 years ago when the first computational models of neurons were created (McCulloch-Pitts Neuron). Recent advancesincluding GPT-4have been powered by neural nets, a type of AI that is loosely based on neurons and includes natural language processing. I would be remiss not to point you to the fantastic article by Stanford Professor Chris Manning, director of the Stanford Artificial Intelligence Laboratory. The first few pages provide a fantastic history leading up to the current models.

You say that computational technologies have struggled with natural language processing and complex or domain-specific tasks like those in the law, but with advancing capabilities of large language modelsand GPT-4you sought to demonstrate the potential in law. Can you talk about language models and how they have improved, specifically for law? If its a learning model, does that mean that the more this technology is used in the legal profession (or the more it takes the bar exam) the better it becomes/more useful it is to the legal profession?

Large language models are advancing at a breathtaking rate. One vivid illustration is the result of the study I worked on with law professors and Stanford CodeX fellows Dan Katz and Michael Bommarito. We found that while GPT-3.5 failed the bar, scoring roughly in the bottom 10th percentile, GPT-4 not only passed but approached 90th percentile. These gains are driven by the scale of the underlying models more than any fine-tuning for law. That is, our experience has been that GPT-4 outperforms smaller models that have been fine-tuned on law. It is also critical from a security standpoint that the general model doesnt retain, much less learn from, the activity and information of attorneys.

What technologies are next and how will they impact the practice of law?

The rate of progress in this area is remarkable. Every day I see or hear about a new version or application. One of the most exciting areas is something called Agentic AI, where the LLMs (large language models) are set up so that they can themselves strategize about how to carry out a task, and then execute on that strategy, evaluating things along the way. For example, you could ask an Agent to arrange transportation for a conference and, without any specific prompting or engineering, it would handle getting a flight (checking multiple airlines if need be) and renting a car. You can imagine applying this to substantive legal tasks (i.e., first I will gather supporting testimony from a deposition, then look through the discovery responses to find further support, etc).

Another area of growth is mutli-modal, where you go beyond text and fold in things like vision. This should enable things like an AI that can comprehend/describe patent figures or compare written testimony with video evidence.

Big law firms have certain advantages and I expect that they would want to maintain those advantages with this sort of evolutionary/learning technology. Do you expect AI to level the field?

Technology like this will definitely level the playing field; indeed, it already is. I expect this technology to at once level and elevate the profession.

So, AI-powered technology such as LLMs can help to close the access to justice gap?

Absolutely. In fact, this might be the most important thing LLMs do in the field of law. The first rule of the Federal Rules of Civil Procedure exhorts the just, speedy and inexpensive resolution of matters. But if you asked most people what three words come to mind when they think about the legal system, speedy and inexpensive are unlikely to be the most common responses. By making attorneys much more efficient, LLMs can help attorneys increase access to justice by empowering them to serve more clients.

Weve read about AIs double-edged sword. Do you have any big concerns? Are we getting close to a Robocop moment?

My view, and the view of Casetext, is that this technology, as powerful as it is, still requires attorney oversight. It is not a robot lawyer, but rather a very powerful tool that enables lawyers to better represent their clients. I think it is important to distinguish between the near term and the long term questions in debates about AI.

The most dramatic commentary you hear (e.g., AI will lead to utopia, AI will lead to human extinction) is about artificial general intelligence (AGI), which most believe to be decades away and not achievable simply by scaling up existing methods. The near term discussion, about how to use the current technology responsibly, is generally more measured and where I think the legal profession should be focused right now.

At a recent workshop we held at CodeXs FutureLaw conference, Professor Larry Lessig raised several near-term concerns around issues like control and access. Law firm managing partners have asked us what this means for associate training; how do you shape the next generation of attorneys in a world where a lot of attorney work can be delegated to AI? These kinds of questions, more than the apocalyptic prophecies, are what occupy my thinking. That said, I am glad we have some folks focused on the longer term implications.

Pablo Arredondo is a Fellow at CodeX The Stanford Center for Legal Informatics and the co-founder of Casetext, a legal AI company. Casetexts CoCounsel platform, powered by GPT-4, assists attorneys in document review, legal research memos, deposition preparation, and contract analysis, among other tasks. Arredondos work at CodeX focuses on civil litigation, with an emphasis on how litigators access and assemble the law. He is a graduate of Stanford Law School, JD 05, and of the University of California at Berkeley.

Read more:
GPT-4 Passes the Bar Exam: What That Means for Artificial ... - Stanford Law School

Pulse Lab Kampala (PLK) has called on stakeholders to support the creation of an environment that fosters the use of data and artificial intelligence (AI).

Speaking at the Conference on the State of Artificial Intelligence in Africa (COSAA) held at Strathmore University in Nairobi, Kenya, last month, Ms Morine Amutorinea data associate at PLKhighlighted the importance of collaboration in creating platforms and communities that can help scale up projects beyond Africa.

When working alone, a project cannot scale beyond Uganda or even Africa at large, noted Ms Amutorine.

During the event, Ms Amutorine showcased a radio mining tool developed by PLK.

The AI-powered social listening tool can monitor multiple radio stations at the same time and filter out content based on specific keywords.

When we wanted to assess public opinion about Covid-19 in Uganda, specific keywords would be fed in and the tool would retrieve audio clips of radio talk shows that were discussing that topic, she explained.

ProwessThe tool can listen to more than 20 radio stations simultaneously and comprehend three local languages in UgandaUgandan English, Acholi, and Luganda.

Mr Martin Gordon Mubangizi, a data scientist doubling as the PLK lead, explained that the Automatic Voice Recognition tool should be trained with a vast dataset of content, including text, pairs of text and audio, a list of all words in a given language and then attribute the rightful pronunciation.

Ms Amutorine toldMonitor that AI does not dispense with human effort.

The latter is needed at the stage of transcribing and analysing the data in order to assess personal perceptions.

Mr Pius Kavuma Mugagga, a data engineer with PLK, consequently demonstrated another AI tool. Named the Pulse Satellite Tool, it is a product of a group of researchers at PLK that was birthed in 2015 in partnership with the United Nations Satellite Centre (UNOSAT). The goal then was to develop an AI tool that would estimate economic development of a region.

We started by experimenting with satellite imagery, Mr Mugagga revealed, adding, The AI would then be programmed to identify, highlight and count shelter roof tops and later the results would be used to assess settlement mapping.

He went on to explain that by looking at the nature of roof tops, you may be able to make some inference regarding the economic transition of a place.This, according to Mugagga, led to the initial conception of the Pulse Satellite Tool that was eventually taken on by UNOSAT. The tool proved handy in flood mapping.

In January 2021, Mozambique experienced floods caused by Tropical Cyclone Eloise. They requested UNOSAT to do a rapid mapping on the area, Mr Mugagga revealed, adding, Using the Pulse Satellite Tool, UNOSAT was able to identify the areas of interest that were assessed for service delivery.

Not ready to rest on its laurels, PLK is already looking to develop a platform where anyone can upload a satellite image whose results would be downloaded for use after the mapping process.

However, there should be legal terms to it for it to be operational outside the UN systems, Mr Mubangizi told Saturday Monitor, adding that that is why Uganda has joined Nigeria in being one of only two countries in Africa to embark on a national data strategy.

Mr Mubangizi further revealed: PLK, in partnership with the Ugandan Ministry of ICT and National Guidance, are looking at achieving such goals by 2040 where big data and AI governance will be accessible by everyone for use, reuse and sharing.

The conferencethe first of its kind in Africa highlighted the potential for AI to transform the culture of the United Nations and deepen its impact. Experts say by creating a supportive environment for data and AI, stakeholders can unlock opportunities for innovation and growth, driving positive change across Africa and beyond.

About artificial intelligence

Since the early days of computers, scientists have strived to create machines that can rival humans in their ability to think, reason and learn in other words, artificial intelligence (AI).While todays AI systems still fall short of that goal, they are starting to perform as well as, and sometimes better than, their creators at certain tasks. Thanks to new techniques that allow machines to learn from enormous sets of data, AI has taken massive leaps forward.

AI is starting to move out of research labs and into the real world. It is having an impact on our lives. There can be little doubt that we are entering the age of AI.

As AI enters the real world by assessing loan applications, informing courtroom decisions or helping to identify patients who should receive treatment, so too does one of its most fundamental flaws: bias.

Algorithms are only as good as the code that governs them and the data used to teach them. Each can carry the watermark of our own preconceptions. Facial recognition software can misclassify black faces or fail to identify women, criminal profiling algorithms have ranked non-whites as higher risk and recruitment tools have scored women lower than men. But with these challenges, there has been mounting pressure on technology giants to fix them.

*Additional information source: BBC

Link:
Africa tipped on power of artificial intelligence - Monitor

M87 supermassive black hole originally imaged by the EHT collaboration in 2019 (left); and new image generated by the PRIMO algorithm using the same data set (right). Credit: Medeiros et al. 2023

Astronomers used machine learning to improve the Event Horizon Telescopes first black hole image, aiding in black hole behavior understanding and testing gravitational theories. The new technique, called PRIMO, has potential applications in various fields, including exoplanets and medicine.

Astronomers have used machine learning to sharpen up the Event Horizon Telescopes first picture of a black hole an exercise that demonstrates the value of artificial intelligence for fine-tuning cosmic observations.

The image should guide scientists as they test their hypotheses about the behavior of black holes, and about the gravitational rules of the road under extreme conditions.

Overview of simulations that were generated for the training set of the PRIMO algorithm. Credit: Medeiros et al. 2023

The EHT image of the supermassive black hole at the center of an elliptical galaxy known as M87, about 55 million light-years from Earth, wowed the science world in 2019. The picture was produced by combining observations from a worldwide array of radio telescopes but gaps in the data meant the picture was incomplete and somewhat fuzzy.

In a study published last week in The Astrophysical Journal Letters, an international team of astronomers described how they filled in the gaps by analyzing more than 30,000 simulated black hole images.

With our new machine learning technique, PRIMO, we were able to achieve the maximum resolution of the current array, study lead author Lia Medeiros of the Institute for Advanced Study said in a news release.

PRIMO slimmed down and sharpened up the EHTs view of the ring of hot material that swirled around the black hole as it fell into the gravitational singularity. That makes for more than just a prettier picture, Medeiros explained.

Since we cannot study black holes up close, the detail of an image plays a critical role in our ability to understand its behavior, she said. The width of the ring in the image is now smaller by about a factor of two, which will be a powerful constraint for our theoretical models and tests of gravity.

The technique developed by Medeiros and her colleagues known as principal-component interferometric modeling, or PRIMO for short analyzes large data sets of training imagery to figure out the likeliest ways to fill in missing data. Its similar to the way AI researchers used an analysis of Ludwig von Beethovens musical works to produce a score for the composers unfinished 10th Symphony.

Tens of thousands of simulated EHT images were fed into the PRIMO model, covering a wide range of structural patterns for the gas swirling into M87s black hole. The simulations that provided the best fit for the available data were blended together to produce a high-fidelity reconstruction of missing data. The resulting image was then reprocessed to match the EHTs actual maximum resolution.

The researchers say the new image should lead to more precise determinations of the mass of M87s black hole and the extent of its event horizon and accretion ring. Those determinations, in turn, could lead to more robust tests of alternative theories relating to black holes and gravity.

The sharper image of M87 is just the start. PRIMO can also be used to sharpen up the Event Horizon Telescopes fuzzy view of Sagittarius A*, the supermassive black hole at the center of our own Milky Way galaxy. And thats not all: The machine learning techniques employed by PRIMO could be applied to much more than black holes. This could have important implications for interferometry, which plays a role in fields from exoplanets to medicine, Medeiros said.

Adapted from an article originally published on Universe Today.

Reference: The Image of the M87 Black Hole Reconstructed with PRIMO by Lia Medeiros, Dimitrios Psaltis, Tod R. Lauer and Feryal zel3, 13 April 2023, The Astrophysical Journal Letters.DOI: 10.3847/2041-8213/acc32d

Go here to see the original:
Artificial Intelligence Reveals a Stunning, High-Definition View of M87's Big Black Hole - SciTechDaily

Artificial intelligence technologies have the potential to reshape how diagnoses are made, if not revolutionize diagnostics in medicine. Already, technologies exist to streamline the diagnostic process and detect illness before a physician can.

But the technology should not be fully relied upon just yet.

The algorithm can be only as good as the data that we give it, said Elke Rundensteiner, professor of computer science and founding director in data science at Worcester Polytechnic Institute.

PHOTO | Courtesy of WPI

Elke Rundensteiner's research at WPI focuses on effective use of data.

Its a mistake to think that technology immediately holds less bias than humans. Artificial intelligence programs arent developed in a vacuum: In medicine, they are fed thousands of data points from practitioners based on real data from real patients, and the AI learns to analyze based on data not always representative of diverse groups.

Participation in clinical trials in medicine is not representative of all populations, said Rundensteiner. According to a 2021 report in the Journal of the American Medical Association, less than 5% of clinical trial participants are Black.

If the data is biased and you give it to an algorithm, then the system might be neutral, but the bias is already baked in, she said.

Dr. Neil Marya, a gastroenterology specialist at UMass Memorial Health and assistant professor at UMass Chan Medical School in Worcester, is researching new diagnostic tools for pancreatic cancer and bile duct cancer. While a few years from clinical use, said Marya, the technologies are showing promise through observational use.

The AI is trained on tens of thousands of image and video data points from patients with and without cancer. The goal is to use computer learning to diagnose cancer, but for now, it is not enough to rely on for a definitive diagnosis.

We're not acting on the AI right now. Its just in the background, and we are understanding how it works in these cases, Marya said. For now, humans outrank the technology in declaring something cancerous or noncancerous. They will not yet use the technology to recommend chemotherapy or surgery without a definitive biopsy, he said.

In some cases, the AI has said something is cancer far before a pathologist has been able to make a diagnosis, Marya said. However, at other times, it has been incorrect, giving a false negative. In either case, the data the AI is pulling from is potentially more useful for certain demographic groups than others.

It would stand to reason to say, AI, should supersede all that implicit bias since its not influenced by one's upbringing or other things. But the issue is that for a lot of these AI models that we're developing, they're being developed at these major academic medical centers in areas or regions where there are only certain types of patients that are getting access to the care that generate the data necessary to create these artificial intelligence models, Marya said.

Maryas original work on the topic of pancreatic cancer models was at the Mayo Clinic in Minnesota, where he said a majority of the patients they saw were white males over the age of 60. Due to the regional bias of Southern Minnesota, there were a very low number of African Americans in the study, he said.

So while the results were very promising, indicating the possibility artificial intelligence programs could be trained to detect pancreatic cancer before physicians could use the current gold-standard method of biopsy, they might not be applicable to other demographic groups.

There's no denying that it's a heavily biased dataset towards a particular type of person. And it might not perform as accurately or maintain these really nice performance parameters when we put it out into practice, Marya said.

This chart shows the increase in clinical trials related to artificial intelligence in the U.S.

While there is cause for a measured introduction of the new machine-learning technology into the practice of medicine, increased access to software physicians can use in tandem with their expertise may increase access to needed health care in less serviced areas, said Dr. Bernardo Bizzo, senior director of the Data Science Office at Mass General Brigham in Boston.

Bizzos team works on software as a medical device, he said. With success, they have developed technology to detect early acute infarcts in some instances that neuroradiologists did not. In his estimation, a wave of new, similarly capable software is here.

We are at an inflection point with this [AI] technology. The expectation is that there will be a lot more AI products made available at a much faster pace, he said.

Marya said he may appear too nervous in his full adoption of the technology, but he just wants to work out the issues. I love the research we're doing; and I'm really passionate about it, and I trust it; but I think we need to be measured with how we approach this, he said. It's really important to do this right.

Rundensteiner said, based on how the public responds to AI overall, patients will continue to want the human element in the forefront, even if the technology is delivering more definitive answers.

Most [patients] would probably still say that they want that human touch right now. But I believe it's just a matter of time until we might want that convenience, and the human is maybe just a face because even that human that they're talking to has computers behind them, she said.

Patients and their families are often less skeptical than expected, said Marya. When it comes to life-threatening illnesses, patients and their families are open to most technologies with minimal hesitation.

If there was a better way, a more accurate way to get a diagnosis sooner, they're all for it, he said. A lot of people, for good and bad, I think they just want the most accurate diagnosis done as soon as possible. And however that's done, I think people just want the best results.

The rest is here:
Diagnosing AI: Healthcare community excited, wary of artificial ... - Worcester Business Journal

Adults have to do a lot of unpleasant jobs; its part of the gig. Taking out the trash, cleaning the toilet and making the daily school run are unavoidable when it comes to keeping life running smoothly.

But theres one particular job that fills me with dread: calling a helpline."

Every time I pick up the phone to discuss tax codes, remortgage rates, insurance quotes, doctors appointments or some other exciting aspect of modern life, my knees go slack and my head starts to pound. Cue generic hold music and a constant robotic reminder of my place in the virtual queue.

Once you do get through to a person, things rarely improve. The poor soul on the other end of the line guides me through mundane security questions before reading from a pre-prepared script. Often, they fail to offer up a single noteworthy piece of advice when questioned directly.

During one of these recent calls, it occurred to me everyone involved would benefit from letting artificial intelligence handle the task. I dont mean the basic interactive voice response (IVR) program that routes your call based on how you answer recorded questions; I mean a full conversational AI agent capable of discussing and actioning my requests with no human input.

Id get through the process faster (because the organization wouldnt need to wait for available humans to assign) and it wouldnt require a living, breathing person to spend their days on the phone to an aggravated person like me. Similarly, an AI doesnt need to clock off at the end of a shift, so the call could be handled any time of the day or night.

Plenty of companies have implemented browser or app-based chat clients but, the fact is, a huge amount of people still prefer to pick up the phone and do things by voice. And I think most industry leaders recognize this.

Humana, a healthcare insurance provider with over 13 million customers, partnered with IBMs Data and AI Expert Labs in 2019 to implement natural language understanding (NLU) software into its call centers to respond to spoken sentences. The machines either rerouted the call to the relevant person or, where necessary, simply provided the information. This came after Humana recognized that 60% of the million-or-so calls they were getting each month were just queries for information.

According to a blog post (opens in new tab) from IBM, The Voice Assistant uses significant speech customization with seven language models and two acoustic models, each targeted to a specific type of user input collected by Humana.

Through speech customization training, the solution achieves an average of 90-95% sentence error rate accuracy level on the significant data inputs. The implementation handles several sub-intents within the major groupings of eligibility, benefits, claims, authorization and referrals, enabling Humana to quickly answer questions that were never answerable before.

The obvious stumbling block for most companies will be the cost. After all, OpenAIs chatbot ChatGPT charges for API access while Metas LLaMA is partially open-source but doesnt permit commercial use.

However, given time, the cost for implementing machine learning solutions will come down. For example, Databricks, a U.S.-based enterprise company recently launched Dolly 2.0 (opens in new tab), a 12-billion parameter model thats completely open source. It will allow companies and organizations to create large language models (LLMs) without having to pay costly API fees to the likes of Microsoft, Google or Meta. With more of these advancements being made, the AI adoption rate for small and medium-sized businesses will (and should) increase.

According to recent research by industry analysts Gartner (opens in new tab), around 10% of so-called agent interactions will be performed by conversational AI by 2026. At present, the number stands at around 1.6%.

"Many organizations are challenged by agent staff shortages and the need to curtail labor expenses, which can represent up to 95 percent of contact center costs, explained Daniel O'Connell, a VP analyst at Gartner. Conversational AI makes agents more efficient and effective, while also improving the customer experience."

You could even make the experience a bit more fun. Imagine if a company got the license to utilize James Earl Jones voice for its call center AI. I could spend a half-hour discussing insurance renewal rates with Darth Vader himself.

I could spend a half-hour discussing insurance renewal rates with Darth Vader himself.

Im not saying there wont be teething problems; AI can struggle with things like regional dialects or slang terms and there are more deep-rooted issues like unconscious bias. And if a company simply opts for a one-size-fits-all AI approach, rather than tailoring it to specific customer requirements, we wont be any better off.

Zooming out for a second, I appreciate that were yet to fully consider all the ethical questions posed by the rapid advancements in AI. Regulation will surely become a factor (if it can keep pace) and upskilling a workforce to become comfortable with the new system will be something for industry leaders and educational institutions to grapple with.

But I still think a good place to start is letting the robots take care of mundane helpline tasks its for the good of humanity.

Today's best Google Nest Hub (2nd Gen) deals

See the original post:
I can't wait for artificial intelligence to take this job away from humans - Tom's Guide