Your AI Update—September 2024
Keep up with recent news in the world of Generative AI, including new features, AI in the workplace, social and ethical implications, regulations, and research revelations. Though typically focused on the most recent month, given our return from summer break, this issue allows a bit longer window (16m read).
Tech Titans: New Features, Products, and More
Beyond Text: Your AI that sees, hears, and speaks
While still “wowing” us with improvements in text generation and image generation, recently, AI companies have started to evolve their product to include sight, sound, and speech.
First announced in May, OpenAI and its new model, GPT 4o, introduced conversational speech, vision capabilities, and live audio recognition to the ChatGPT product line. In essence, the model can be conversed with, describe its surroundings (only if the camera feature is turned on), and seamlessly combine its speech, vision, and language generation features. These capabilities, later matched by Google’s Gemini 2, open the door for a wide range of possibilities and use cases beyond sentence generation. For example, using the ChatGPT app with the paid version, you can talk to the app in one language and have it instantly answer in another. Using the vision features, you can take a picture of a meeting and it will describe the scene in remarkable detail, including recognition of the content on a projection screen, comments about the social aspects of the meeting, and the formality of the dress code. Putting all these features together, one can imagine complex social and academic support systems reaching into interactional and physical dimensions beyond text. Not sure how to fix a broken door? Take a picture and ChatGPT may diagnose it. Not sure how to solve a mathematics problem? Take a picture and ChatGPT will teach you. The text-based world of the Large Language Model (LLMs) is being surpassed by this new approach called the Large Multi-Modal Model (LMM).
Given that our day-to-day activities are not confined to text-based activities, these changes have the potential to dramatically expand the usefulness of these new systems. Considering Apple is in discussions with Open AI, and Google controls both Gemini and the Android operating system, ubiquitous access to these systems may soon be on the horizon.
Apple finally RSVP’d to the Gen AI Party
After months of speculation and anticipation, Apple CEO, Tim Cook, announced Apple will be launching a new AI approach called Apple Intelligence. With their extensive and loyal base of users, Apple is priming the world for a new level of AI literacy or at the very minimum, AI integration into daily life.
With “Apple Intelligence”, Apple promises to create a more conversational Siri, implement generative writing tools, and provide AI-integrated photo editing while maintaining their commitment to highly secure personal data. These features will debut later this fall and will only be available for the newest Apple devices due to the necessary hardware upgrades. Apple’s push for AI-integrated devices, a move that was already matched by Samsung and Google, marks the likely beginning of an AI-integrated future: that is if they follow through with their promises.
And then there were four (or 5?)
While most people are well aware of ChatGPT and its creator, OpenAI, many remain unaware that there are a few other prominent systems on the market getting wide use, adoption, and discussion:
-
OpenAI’s ChatGPT: Mostly stand-alone but available in various configurations and compatible with other systems via APIs.
-
Microsoft Co-pilot: Built with pieces including OpenAI’s ChatGPT technologies, the Co-pilot suite is being infused into the full range of common Microsoft products and is available for technologists through Microsoft's Azure cloud platforms.
-
Google Gemini: Similar to Microsoft, Google is embedding Gemini capabilities into the range of their products including search and Android operating systems. They are also happy to sell technologists access via their Google Cloud offerings.
-
Anthropic’s Claude products: Anthropic is well known among early adopters for producing standout writing/coding abilities while maintaining perhaps the industry’s strongest customer safety and ethics stance. Their actions in protecting data and sharing system design details have forced others to up their game—even their free versions do not use your data for training by default.
-
Meta’s Llama open source tools. Meta does not sell access to its Llama technologies but gives it to technologists for free and embeds it in their products such as Facebook and Instagram. This means it is being widely used in the back of many AI systems and has an important role in the evolving market.
No, we didn’t forget
In the last month, OpenAI released a new underlying model called o1 (pronounced “oh one”) specially designed for improved reasoning and multi-step processing. According to OpenAI’s release statement on September 12th, o1 is the first in a new suite of models “designed to spend more time thinking before they respond. They can reason through complex tasks and solve harder problems than previous models in science, coding, and math.” As this model is still only available in “Preview,” users and researchers are still digesting the possibilities and assessing the various claims—we’ll have more to say in the next update.
Read more about…
-
…how AI companies are influencing journalism: OpenAI partners with Times Magazine
-
…taking on the tech giants: OpenAI and Google workers seek protection for whistle-blowing
AI at Work: Work, Commerce, the Economy and Labor
Tech Skills are Not Enough in the AI Gold Rush
While a job in the tech industry previously guaranteed job security and a high-paying salary, many tech workers like software engineers and software developers have struggled to retain employment.
In 2024 alone, 137,500 workers have been laid off from nearly 500 tech companies. While AI is not the only factor causing the recent spike in layoffs, the need for upskilling and AI-targeted hiring has become more prevalent. A recent McKinsey study argued that by 2030, 14% of employees will have been forced to change their careers because of AI as tech giants’ investment in artificial intelligence continues to increase. In the last quarter alone, Microsoft, Google, Meta, and Amazon cumulatively invested $52.9 billion in the technology, often sourcing these funds from non-AI workforce reductions.
An analysis of recent job listings confirms that AI skills are highly desirable in the current job landscape; while AI-related job listings are rising year-over-year, tech listings are falling. This raises the question: What are these AI jobs, and how does one remain hireable in the age of AI? The majority of these AI-related job listings require generative AI skills, and many employers prefer candidates with knowledge of platforms like ChatGPT, experience with prompt engineering, and a baseline understanding of how AI is trained.
The future of AI in the workplace is not all doom and gloom. According to the Council of Economic Advisors, even though AI could negatively impact some workers due to declining demand, it is unlikely to impact overall employment. Alorica, a California company that runs customer service centers, illustrates how this might look. The company recently implemented a translation tool that lets its representatives talk with customers who speak 200 different languages. With this increased employee productivity, the company is still hiring aggressively. Examples like these instill hope that AI might turn out to be “like breakthroughs of the past”; that is, even though it may eliminate some jobs it will simultaneously create others.
Shaping Business Strategies: AI for Big and Small Companies
With its promise of streamlined efficiency, high return on investment, and increased workflow, businesses across several industries have hopped on the train and introduced AI into their work cycle. It is without surprise that one of the world’s largest retailers, Amazon, recently announced how they use their home-grown AI system, Amazon Bedrock, to increase sales. Amazon Bedrock relies on a variety of foundation models (FMs) including those provided by Meta, Stability AI, and most prominently, Anthropic’s Claude AI, aiming to provide a more personalized customer experience. In a survey recently conducted by KPMG, it was reported that from a group of 225 C-Suite and senior business leaders representing organizations with annual revenue of $1 billion or more, 52% believe AI opens opportunities for more revenue. As reported by the U.S. Chamber of Commerce, this landscape of opportunity looks similar for small businesses. Between 2023 and 2024, there was a 17% increase (from 23% to 40%) in small companies that have reported using AI in areas such as payroll management, promotions, accounting, and processing sales. While some AI services are expensive to implement, these companies reported taking advantage of free services, including ChatGPT, for tasks that do not require as much tailored expertise. When it comes to online platforms for job seekers and employers, LinkedIn is an example of a company that has incorporated AI for both employers and premium subscribers. Their LLMs are trained with user data and their models range from helping professionals explore career opportunities to even personalizing cover letters. A survey conducted by McKinsey’s Global Institute reported that around half of the companies that work in labor management use AI at a large scale, and by the year 2030 it is estimated that number will rise to 70%.
The widespread adoption of these technologies defines a new axis of competition among business strategies. As companies continue to recognize the potential of AI, failure to adapt to it could potentially lead to difficulty keeping pace in a data-driven economy. It would appear that companies that embrace AI’s potential are better equipped to navigate the demands of a rapidly evolving market. Integrating these technologies can help companies streamline skill development and foster an environment of learning in an ever-changing professional landscape.
Read more about…
-
…how retailers are experimenting with Gen AI: How to scale gen AI in retail
-
…AI in financial market analyses: How Palmyra-Fin is Redefining Market Analysis
AI in the World: Shaping Families, Friends, Lifestyles and Society
AI and the 2024 Election: Revolutionary Force or Overhyped Distraction?
As the 2024 presidential election continues to unfold, many Americans are concerned about the role of AI in transforming political campaigns and voter behavior. While some argue that artificial intelligence influences how electorates perceive candidates via the spread of AI-generated memes and deep fakes, others say that its impact is being exaggerated and that AI's contributions are not fundamentally changing presidential campaigning or voter behavior. As the debate continues, the question remains: is AI revolutionizing the election landscape, or is its influence being overstated?
Camp 1: The Sky is Falling
On one hand, many believe the spread of political misinformation via AI has bred distrust in American voters. A new survey found that 43% of respondents believe that AI will make it much or somewhat more difficult to find accurate information on the presidential election. Most recently, Taylor Swift was forced to respond after AI-generated pictures falsely claiming her endorsement of one of the presidential candidates surfaced online. This incident comes after deep fake robocalls mimicking President Biden and encouraging citizens not to vote in the Democratic Primary spread throughout New Hampshire earlier this year. These are just a couple of examples of the growing challenge that comes with navigating a political landscape increasingly shaped by AI-generated content.
Camp 2: The Sky is… going to be fine
Despite these well-known examples and the depth of suspicion in the general population, some security experts argue that AI’s impact on the 2024 election is being severely overblown. Cait Conley, the senior adviser to the director of the Cybersecurity and Infrastructure Security Agency, asserts that “generative AI is not going to fundamentally introduce new threats to this election cycle.” While it’s undeniable that some have tried to use AI to influence voter opinions, there is no data to suggest that these attempts have been at all successful. Mass persuasion, especially as it applies to complex political beliefs that are influenced by a multitude of mediums, is “notoriously challenging” and emerging research challenges the idea that AI is sophisticated enough to achieve it. Furthermore, the “tsunami of information” published daily by journalists, social media users, political campaigns, and news organizations makes it unlikely for targeted AI disinformation to peek above the surface and have any mass effect.
As November 5th rolls around, the true impact of AI on the democratic process remains uncertain. While the technology has certainly introduced new intricacies that must be navigated, its actual influence on voter behavior and election outcomes is still up for debate.
Redefining the Classroom: The Evolving Role of AI in Education
For years, AI has had little or no place in the classroom. However, this stance has started to shift at all levels of education.
In 2023, 55% of surveyed K-12 state education officials expressed interest in guidance on AI policy development–a figure that jumped to 90% in 2024. At universities, 72% of faculty reported using generative AI for at least one instructional purpose. With artificial intelligence stepping into a more prominent role in education, schools and technology companies alike are responding to this growing demand.
As AI becomes more prevalent at universities, innovative experiments are pushing the boundaries of how these tools might complement–or even challenge–the traditional roles of educators. This year, David Game College in London is experimenting with a “teacherless” course taught by artificial intelligence instead of human educators. The students participating in the pilot program will use virtual reality headsets and the Sabrewing AI model to access a personalized learning experience that assesses their strengths and weaknesses to inform their learning plan. AI will not be left completely alone in the classroom; “learning coaches” will assist in monitoring the course and in teaching disciplines like art, which the administration maintains AI is not capable of teaching. Nonetheless, many critics argue that this approach to learning will “dehumanize the learning process” and will harm students' ability to form interpersonal relationships.
While David Game College has adopted a radical approach to using AI to replace traditional teaching, most companies that develop AI technologies are taking a more practical path to meeting the demands of the education industry. Recognizing that AI cannot replicate the social and emotional development teachers foster, these companies have focused on developing tools that support, rather than replace, educators. Most recently, Google’s Gemini has released a suite of tools aimed at helping both students and teachers make the most of AI's capabilities. For example, learners who chat with Gemini via their school accounts now know their data isn’t being used to train future AI models, and educators have access to digital skills lessons that teach students the fundamentals of up-and-coming technologies.
Read more about…
-
…AI in sports betting: AI reveals NFL against the spread, over-under picks for Week 1
-
…parents’ concern over how teens use AI: Parents have no idea how teens are using AI
Taming AI: Ethics, Policies and Regulations
Because it is nearly impossible for a large company to operate in the US without obeying California laws, when it comes to regulation, technology companies keep their eyes on California. Likewise, most organizations working or selling outside of the US keep their eyes on the EU (the world’s third-largest economy), given its breadth and regulatory enthusiasm. In both cases, legislatures are attempting to thread the needle between protecting their citizens and hindering innovation and economic growth.
California’s Divisive Proposed AI Safety Bill
While the world tries to find the balance between AI regulation and innovation, all eyes are on California, where the major tech companies are waiting to hear what laws they will have to comply with soon.
The legislature in California passed a proposed AI safety bill on August 29, 2024. California’s governor then had until September 30, 2024, whether to sign it into law. If he chose to sign it, “the California attorney general [would] then have the power to stop AI firms from building AI models if the courts find them dangerous.” Governor Newsom had a difficult choice to make, and it was made more complicated by the many actors who have weighed into the public conversation surrounding the bill.
The controversial bill was opposed by powerful figures in the U.S. Congress, such as Nancy Pelosi, who cited fears of limiting Silicon Valley’s innovation as the reason for her dissent. It was also notably opposed by many “tech giants” such as Google, Meta, and OpenAI. However, it had support from Anthropic, a major AI player that frequently touts its commitment to ethical AI development. Anthropic even suggested edits that made it into the version of the proposed law which was passed by the State’s Senate.
On September 29, 2024, Governor Newsom vetoed the bill, asserting that it was “‘well-intentioned’ but left out key ways of measuring risk and other consumer harms.” Newsom continues to lightly tip-toe the line between regulator and friend to Silicon Valley. This veto came after a long line of approvals for smaller AI safety initiatives from the governor’s office, including measures towards “cracking down on deepfakes, requiring AI watermarking, protecting children and workers, and combating AI-generated misinformation.” In this way, Newsom was able to signal his value of the safe use of AI, but he also avoided placing intensive limitations on the major tech players in his state.
What the state of California does in terms of AI regulation is especially important due to Silicon Valley’s positioning within the state. Additionally, since few regulations exist at the state or federal level at the moment, any legislation passed now or in the near future will set a strong precedent for future laws surrounding AI safety.
The European AI Safety Agenda
The European Union has been a trailblazer in AI regulation, so watching its actions could reveal major trends to look out for in the U.S. and other contexts.
The European Union’s AI Act went into effect on August 1, 2024. Importantly, “the Act has broad extraterritorial implications, extending its reach to providers who place or put AI systems into service within the EU market, regardless of their location.” So, this act will change regulatory measures not just for companies based in the EU, but for many more around the world. This is likely to cause friction between the regulatory body and companies within the AI realm—we have recently seen Meta pause their plans for AI in the EU due to strict regulations, even before this law went into effect. Such hesitancy from a tech superpower exemplifies the fears of legislators across the world who want to ensure the safety of their citizens, but not at the expense of tech business.
The regulations are based on the perceived risk of the AI, with systems being labeled as unacceptable, high, limited, or minimal risk. Unacceptable systems are those which are manipulative or exploitative in any number of ways. Higher-risk applications of AI will have more to watch out for than lower-risk ones, as companies that do not meet the new rules will be fined and made to comply.
International cooperation on AI goes further than just the EU. On September 5, 2024, the Council of Europe opened for signing a global treaty on the safe and responsible use of AI. It is “the first-ever international legally binding treaty aimed at ensuring that the use of AI systems is fully consistent with human rights, democracy and the rule of law.” The official name of the treaty is the Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law. It is meant to enumerate signatory countries’ commitment to using AI in a productive way that respects human rights—it even references the 1948 Universal Declaration of Human Rights along with other conventions on human rights. Current signatories include the U.S., the U.K., the EU, Israel, Norway, Iceland, Andorra, Georgia, the Republic of Moldova, and San Marino.
The specific articles of the convention range from non-discrimination to data privacy to oversight mechanisms, and each member of this conference will report to the rest within two years on how they have implemented the parts of this treaty. Interestingly, measures to use AI for the purpose of national defense and security are not required to apply to the rules of this convention.
Read more about…
-
…Colorado’s Groundbreaking AI Legislation: Consumer Protections for Artificial Intelligence
-
…the US Government re-committing to AI Safety: US government re-launches AI safety tool
Research Revelations
Data Transparency Research Reveals AI’s New Consent Crisis
Large Language Models like those underlying ChatGPT and similar systems learn how language works by ingesting millions of documents and quantifying patterns in human language. The quality, scope, and origin of these data have a very significant impact on the performance and appropriateness of the chat system’s behavior. However, the documentation for these datasets, including licensing and source attribution, is often incomplete or incorrect, raising the issue of how AI companies can ethically source their data. To address this problem, a team from MIT and other institutions audited over 1,800 datasets and discovered that over 70% lacked clear licensing information, and many had incorrect or missing attributions. They developed a tool called the "Data Provenance Explorer" to trace the origins and licenses of datasets, allowing AI developers to (1) make better decisions when selecting data, (2) ensure they meet legal and ethical standards, and (3) improve the quality of the models they create.
One surprising finding this initiative revealed is that access to data is diminishing. Many websites have started restricting AI companies from scraping their content. One study looked at 14,000 web domains (you can think of them as websites) that are included in three commonly used AI training data sets and found that up to 25% of high-quality data in commonly used datasets is now restricted by sites using the Robots Exclusion Protocol, a decades-old method for website owners to restrict automated bots.
Consequently, a "consent crisis" arises, as publishers and content creators feel more and more like their data is being used without permission or compensation. While large AI companies can strike deals with publishers, smaller companies and researchers struggle to access crucial training data. This problem calls for new ethical tools and regulations to not only allow the underdogs to compete but also balance the interests of both data owners and AI developers.
Unlocking the Brain: How AI is Revolutionizing Movement
Due to the innate complexity and overlapping nature of the signals our brains generate, decoding brain patterns has long been a challenge for scientists. Each thought, movement, or cognitive state creates electrical activity that occurs simultaneously with other processes. However, new developments in the medical field of AI are looking to help untangle this enduring problem and allow neurologists to gain full insights into brain activity.
Maryam Shanechi and her team at USC have developed an advanced AI algorithm called DPAD (Dissociative Prioritized Analysis of Dynamics) that can separate and decode specific brain patterns associated with behaviors, like arm movement for example, from other concurrent brain activities. DPAD battles the complexity of these brain signals by prioritizing the learning of behavior-related patterns first, preventing them from being masked by other brain activities. This is crucial for improving brain-computer interfaces (BCIs), which aim to help paralyzed patients regain movement by interpreting their brain signals and translating them into actions, such as controlling a robotic arm. Beyond movement, DPAD has potential future applications in decoding mental states as well – creating the possibility of personalized treatment for mental health conditions by tracking symptom states in real-time.
This research builds upon growing brain-to-speech technology and non-invasive electroencephalography (EEG) to capture brain signals, where potential already exists to give a voice back to individuals with conditions like ALS or other speech-impairing disabilities. In an impressive clinical trial conducted by Dr. Leigh Hochberg and his research team BrainGate, a new BCI made significant strides in brain-to-speech technology. During the 84-session study, the system translated brain signals into speech with up to 97% accuracy, enabling a man with ALS to communicate using a voice that mimicked his pre-disease speech. These advancements open the door for a world of opportunities to better support both physically and cognitively disabled people. ~
Read more about…
-
…AI’s impact on research & scientific publishing: How is AI Changing Scientific Publishing?
-
…Roblox's Launching of a New Generative AI to Make Scenes: Roblox’s Road to Generative AI
***all imagery created using Image Creator from Designer***
The New AI Project | University of Notre Dame
Editor: Graham Wolfe
Contributors: Clare Hill, Aiden Gilroy, Mary Claire Anderson, Annie Zhao, Gaby Sanchez
Advisor: John Behrens
Recommended Videos
Ted Talks
Sal Khan: How AI could save (not destroy) education
Ray Kurzweil: The last 6 decades of AI — and what comes next
What is AI?
The National Science Foundation explains: what is AI?
Microsoft AI CEO Mustafa Suleyman predicts the future of AI.
Archive