While autonomous agents often surpass humans in their ability to handle vast and complex data, their potential misalignment (i.e., lack of transparency regarding their true objective) has thus far hindered their use in critical applications such as social decision processes. More importantly, existing alignment methods provide no formal guarantees on the safety of such models. Drawing from utility and social choice theory, we provide a novel quantitative definition of alignment in the context of social decision-making. Building on this definition, we introduce probably approximately aligned (i.e., near-optimal) policies, and we derive a sufficient condition for their existence. Lastly, recognizing the practical difficulty of satisfying this condition, we introduce the relaxed concept of safe (i.e., nondestructive) policies, and we propose a simple yet robust method to safeguard the black-box policy of any autonomous agent, ensuring all its actions are verifiably safe for the society.

RESTUCCIA: I think "smart hospital" may be a misnomer; "patient-centric" or "technologically advanced" is probably more accurate. Our perspective in the early design for our newest facility was, "How can we develop and build a hospital that can leverage technology to ensure the best care is delivered in the most effective and efficient manner?" Flexibility was a big part of that; having the cabling in place, for example, so that every room is both med-surg and ICU-ready. DALE: It started about 12 years ago, when we decided to move to an electronic health record. That system is a source of so much rich data and has become foundational for how we've set up technology across the enterprise.

Tel Aviv-based Run:ai, a startup that makes it easier for developers and operations teams to manage and optimize their AI infrastructure, today announced that it has raised a $75 million Series C funding round led by Tiger Global Management and Insight partners, which also led the company's $30 million Series B round in 2021. Previous investors TLV Partners and S Capital VC also participated in this round, which brings Run:ai's total funding to $118 million. Run:ai's Atlas platform helps its users virtualize and orchestrate their AI workloads with a focus on optimizing their GPU resources, no matter whether they are on-premises or in the cloud. It abstracts all of this hardware away, while developers can still interact with the pooled resources through standard tools like Jupyter notebooks and IT teams can get better insights into how these resources are being used. The new round comes at a time of fast growth for the company.

Did you miss a session at the Data Summit? While interest in AI remains high among enterprise organizations, particularly for its potential to improve decision-making and automate repetitive tasks, many of these businesses are struggling to deploy AI into production. In a February survey from IDC, only a third of companies claimed that their entire organization was benefitting from an enterprise-wide AI strategy. The same poll found that 69% of companies hadn't yet reached production with AI, and instead remained in the experimentation, evaluation, or prototyping phases. The challenges vary from organization to organization, but some common themes include infrastructure and data.

Run:ai, the company simplifying AI infrastructure orchestration and management, today announced that it has raised $75M in Series C round led by Tiger Global Management and Insight Partners, who led the previous Series B round. The round includes the participation of additional existing investors, TLV Partners and S Capital VC, bringing the total funding raised to date to $118M. Run:ai has grown sharply, with a 9x increase in Annual Recurring Revenue in the last year, while the company's staff more than tripled over the same period. The company plans to use the investment to further grow its global teams and will also be considering strategic acquisitions as it develops and enhances the company's Atlas software platform. Omri Geller, Run:ai CEO and co-founder, said, "It may sound dramatic, but AI is really the next phase of humanity's development. When we founded Run:ai, our vision was to build the de-facto foundational layer for running any AI workload. Our growth has been phenomenal, and this investment is a vote of confidence in our path. Run:ai is enabling organizations to orchestrate all stages of their AI work at scale, so companies can begin their AI journey and innovate faster."

Artificial Intelligence is at the core of the Metaverse, but how trustworthy is it? Although there have been many previous studies into AI's trustworthiness in Web 2.0, these cannot be extended to the metaverse, which requires more complicated metrics to assess system performance and user experience. A recent study from a multinational team of researchers suggests that as we currently lack a set of tested trustworthiness metrics, we should not put our trust in AI to run the metaverse [1]. Today's large scale AI integration means that AI has access to vast amounts of user data; AI can leverage the data to uncover sensitive user behavior like visits to certain websites or personal buying habits. To address these concerns, many agencies, corporations, and government bodies have studied Trustworthy AI (TAI), including the European Commission, United States Department of Defense, and FAANG companies (Meta (Facebook), Amazon, Netflix; and Alphabet (Google)).

The lack of time affects both preflight training for, and in-flight operation of, the experiment. This difficulty with time is currently true with the Space Shuttle Program and will persist with the advent of Space Station Freedom operations. Another key factor in space experimentation is the use of fixed experiment protocols. This major constraint severely limits the ability of an earthbound scientist to change the course of an experiment even when the data and current situation clearly indicate that it would be scientifically more valuable to do so. The goal is to help the astronaut become a scientific collaborator with the ground-based principal investigator who designed the experiment.

Unlike other entrants that fashioned good softbot teams from a battery of relatively wellunderstood robotics techniques, our goal was to see if it was even possible to use evolutionary computation to develop high-level soccer behaviors that were competitive with the human-crafted strategies of other teams. Although evolutionary computation has been successful in many fields, evolving a computer algorithm has proven challenging, especially in a domain such as robot soccer. Our approach was to evolve a population of teams of Lisp s-expression algorithms, evaluating each team by attaching its algorithms to robot players and trying them out in the simulator. Early experiments tested individual players, but ultimately, the final runs pitted whole teams against each other using coevolution. After evaluation, a team's fitness assessment was based on its success relative to its opponent.

Our example training script (mnist_mlp.R) trains a Keras model to recognize MNIST digits. To train a model with tfruns, just use the training_run() function in place of the source() function to execute your R script. The metrics and output of each run are automatically captured within a run directory which is unique for each run that you initiate. You can call the latest_run() function to view the results of the last run (including the path to the run directory which stores all of the run's output): The run directory used in the example above is "runs/2017-10-02T14-23-38Z".