It's all about AI!
We start with whether use of AI needs to be controlled in war. We move to see if AI is becoming more costly & beating physicians. Finally, we end with how AI is being used in wars and tax auditing.
Time to stop deployment of AI in weapons for military use?
Civilian and military authorities have expressed concerns regarding the risks associated with autonomous weapons lacking human oversight.
A gathering convened in Vienna aimed to amplify the voices of those apprehensive about the potential application of Silicon Valley's disruptive ethos to the realm of industrial warfare.
Autonomous weapon systems are already seeing widespread deployment in conflict zones such as Ukraine and Gaza.
How drones are revolutionising war
Leveraging a blend of big data and drone technology, these systems assist military strategists in target selection.
Certain types of loitering munitions can be programmed to engage targets autonomously, eliminating the need for direct human intervention.
This shift towards delegating life-and-death decisions to machines has prompted a myriad of ethical, legal, and technological dilemmas.
Suggestions vary regarding how to address this issue: some advocate for the drafting of a new treaty akin to those prohibiting chemical or nuclear weapons, while others argue that existing trade regulations and humanitarian laws are adequate.
Nevertheless, there is widespread acknowledgment of the imperative for human accountability in this domain.
Will AI be more expensive in the future?
After approximately 18 months of fervent activity in the realm of generative AI, major tech enterprises are demonstrating the substantial revenue potential of artificial intelligence. However, concurrently, they are grappling with significant financial investments required to sustain these endeavors.
Microsoft and Google have disclosed remarkable growth in cloud revenue in their recent quarterly reports, attributed to heightened expenditure by corporate clientele on AI services. Meta Platforms Inc., although trailing in the monetization of AI, has underscored the positive impact of its AI initiatives on enhancing user engagement and refining advertising targeting strategies.
Microsoft, in its announcement on April 25, revealed a capital expenditure of $14 billion for the most recent quarter, anticipating a significant escalation in such costs, partly driven by investments in AI infrastructure. This marked a 79% surge compared to the corresponding quarter of the preceding year.
Alphabet reported a $12 billion expenditure for the quarter, reflecting a 91% increase from the previous year, and anticipates maintaining or surpassing this level throughout the year, focusing on AI-related opportunities.
Meta revised its investment projections for the year, estimating capital expenditures ranging between $35 billion and $40 billion, representing a potential 42% increase at the upper end of the spectrum, citing robust investment in AI research and product advancement.
It has long been acknowledged that the costs associated with AI are on an upward trajectory due to two primary factors: the burgeoning size and cost of developing AI models, and the global demand for AI services necessitating the expansion of data center infrastructure.
Presently, the prevailing cost for training AI models in the market stands at approximately $100 million. Forecasts suggest that forthcoming AI models, slated for release later this year or early next year, may incur costs closer to $1 billion. Moreover, projections for 2025 and 2026 indicate a further escalation in costs, potentially reaching $5 to $10 billion.
A significant portion of these expenses is attributed to chip procurement. AI enterprises heavily rely on graphics processing units (GPUs) to train large language models efficiently. However, the scarcity and steep pricing of these chips, particularly those incorporating cutting-edge features predominantly supplied by Nvidia Corp., pose challenges.
Concurrently, there is a race among major tech players, including Meta, Amazon, Microsoft, and Google, along with other cloud computing providers, to expand their data center footprint. These facilities, often tailored to specific requirements, accommodate arrays of hardware components, cooling systems, and electrical infrastructure.
Projections indicate a substantial surge in spending on data center construction and outfitting, with an anticipated expenditure of $294 billion in 2024, up from $193 billion in 2020.
While chip procurement and data center infrastructure constitute significant portions of AI-related expenditures, some companies are also investing substantial sums in data licensing agreements with content providers.
OpenAI has entered into agreements with several European publishers to integrate their news content into ChatGPT and facilitate AI model training.
Google reportedly secured a $60 million data licensing deal with Reddit, as reported by Reuters.
Additionally, there have been discussions within Meta about potentially acquiring book publisher Simon & Schuster, as reported by The New York Times.
The drive towards larger AI models, along with the increased need for chips and data centers to support their development, and data licensing agreements are escalating expenses for tech companies.
AI outperforming physicians in clinical reasoning processes?
A study conducted by researchers at Beth Israel Deaconess Medical Center revealed that the generative artificial intelligence tool, ChatGPT-4, exhibited superior performance compared to hospital physicians and residents in various aspects of the clinical reasoning process.
However, it was noted that ChatGPT-4 did not outperform human counterparts in all scenarios, prompting discussions on the integration of AI in healthcare settings.
Despite demonstrating higher scores in clinical reasoning, ChatGPT-4 displayed a tendency to make incorrect diagnoses. For instance, an instance highlighted the AI's attempt to administer a pregnancy test to a 74-year-old woman experiencing abdominal pain, a situation where physicians typically would not consider pregnancy as a likely cause.
Despite these limitations, the researchers maintain optimism regarding the potential of AI in healthcare.
Following Institutional Review Board (IRB) approval, the research team distributed case surveys to assess physicians' clinical reasoning skills in July. Subsequent analysis of data comparing physician responses with those generated by ChatGPT-4 occurred in October 2023, with findings published in February 2024.
Using AI during wars
In the summer of 2022, amidst Russia's need for a crucial bridge to facilitate troop resupply west of the Dnipro River, strategic considerations regarding the potential impact of destroying the bridge were meticulously assessed. Central to these deliberations was an inquiry into whether such an action would induce panic among Russian soldiers or their families, along with a strategic focus on maximizing the morale blow through the creation of a targeted "information environment" by Ukraine's government.
The Open Minds Institute (OMI) based in Kyiv played a pivotal role in this endeavor, leveraging advanced artificial intelligence (AI) capabilities to generate comprehensive assessments.
Through sophisticated algorithms, OMI's research outfit meticulously sifted through vast volumes of Russian social media content and socioeconomic data, encompassing diverse metrics such as alcohol consumption, population dynamics, online activity, and consumer patterns.
This AI-driven analysis enabled correlations between evolving sentiments among Russian "loyalists" and liberals regarding the potential consequences for their country's soldiers.
Innovative applications of AI extend beyond psychological warfare, as highlighted by a Ukrainian colonel involved in arms development.
ChatGPT, a renowned AI tool, serves as a foundational resource for drone designers seeking engineering inspiration, particularly in devising strategies to mitigate vulnerabilities to Russian jamming techniques.
Furthermore, AI technology plays a pivotal role in target identification, leveraging extensive data ingestion and analysis of images and textual data to deduce the likely locations of military assets or troop formations.
SemanticForce, a prominent firm with operations in Kyiv and Ternopil, specializes in developing AI models adept at analyzing textual and visual content in response to specific queries.
Leveraging data from various sources, including social media and drone footage, these models discern pertinent information, aiding military intelligence efforts.
Notably, AI capabilities also support Ukraine's counterintelligence endeavors, facilitating the identification of individuals susceptible to betrayal, as emphasized by Oleksiy Danilov, former secretary of the National Security and Defense Council (NSDC).
Ukraine's AI-driven initiatives benefit significantly from widespread societal support, exemplified by citizens' active contribution of data for national defense purposes. Initiatives such as the Diia app, enabling citizens to upload geotagged photos relevant to defense efforts, underscore public engagement.
Additionally, businesses collaborate by providing data to entities like Mantis Analytics, based in Lviv, encompassing diverse information ranging from logistics metrics to security alerts.
Despite the substantial strides in leveraging AI for military and intelligence purposes, the ultimate efficacy of these initiatives remains a subject of ongoing evaluation and debate.
How US is using AI for tax auditing
The New York State Department of Taxation and Finance has reported a significant increase in audits for the year 2022, totaling 771,000 cases, representing a substantial 56% surge compared to the previous year.
This surge in audits comes despite a 5% reduction in the number of auditors employed by New York, now standing at under 200, largely attributed to budget constraints.
A notable aspect of this increased audit activity is the utilization of Artificial Intelligence (AI) technology, which has enabled the state to audit a larger number of taxpayers despite the decrease in human auditors.
This strategic incorporation of AI has allowed New York to maintain its audit effectiveness even with limited manpower resources.
One of the driving factors behind the intensified scrutiny on taxpayers is the relocation trend observed during the Covid-19 pandemic, particularly among high-income individuals who migrated from high-tax states such as California, New York, New Jersey, and Connecticut to states with lower tax burdens like Florida and Texas.
This relocation has prompted states to closely examine the legitimacy and permanency of such moves, especially concerning the tax implications.
In response to this challenge, state tax authorities, in conjunction with AI systems, have adopted innovative approaches, such as analyzing cellphone records, to ascertain the primary residency and tax obligations of individuals.
By scrutinizing data such as cell phone usage patterns to determine where taxpayers predominantly reside and spend their time, tax auditors aim to ensure compliance with tax laws and regulations, regardless of physical location.