8 AI Business Trends in 2024, According to Stanford Researchers

4 months ago 67

AI makes workers much productive, but we are inactive lacking successful regulations, according to caller research. The 2024 AI Index Report, published by the Stanford University Human-Centered Artificial Intelligence institute, has uncovered the apical 8 AI trends for businesses, including however the exertion inactive does not champion the quality encephalon connected each task.

TechRepublic digs into the concern implications of these takeaways, with penetration from study co-authors Robi Rahman and Anka Reuel.

SEE: Top 5 AI Trends to Watch successful 2024

1. Humans inactive outperform AI connected galore tasks

According to the research, AI is inactive not arsenic bully arsenic humans astatine the analyzable tasks of advanced-level mathematical occupation solving, ocular commonsense reasoning and readying (Figure A). To gully this conclusion, models were compared to quality benchmarks successful galore antithetic concern functions, including coding, agent-based behaviour, reasoning and reinforcement learning.

Figure A

Performance of AI models successful antithetic tasks comparative to humans.

While AI did surpass quality capabilities successful representation classification, ocular reasoning and English understanding, the effect shows determination is imaginable for businesses to utilise AI for tasks wherever quality unit would really execute better. Many businesses are already concerned astir the consequences of over-reliance connected AI products.

2. State-of-the-art AI models are getting much expensive

The AI Index reports that OpenAI’s GPT-4 and Google’s Gemini Ultra outgo astir $78 cardinal and $191 cardinal to bid successful 2023, respectively (Figure B). Data idiosyncratic Rahman told TechRepublic successful an email: “At existent maturation rates, frontier AI models volition outgo astir $5 cardinal to $10 cardinal successful 2026, astatine which constituent precise fewer companies volition beryllium capable to spend these grooming runs.”

Figure B

Training costs of AI models, 2017 to 2023.

In October 2023, the Wall Street Journal published that Google, Microsoft and different large tech players were struggling to monetize their generative AI products owed to the monolithic costs associated with moving them. There is simply a hazard that, if the champion technologies go truthful costly that they are solely accessible to ample corporations, their vantage implicit SMBs could summation disproportionately. This was flagged by the World Economic Forum backmost successful 2018.

However, Rahman highlighted that galore of the champion AI models are unfastened root and frankincense disposable to businesses of each budgets, truthful the exertion should not widen immoderate gap. He told TechRepublic: “Open-source and closed-source AI models are increasing astatine the aforesaid rate. One of the largest tech companies, Meta, is open-sourcing each of their models, truthful radical who cannot spend to bid the largest models themselves tin conscionable download theirs.”

3. AI increases productivity and enactment quality

Through evaluating a fig of existing studies, the Stanford researchers concluded that AI enables workers to implicit tasks much rapidly and improves the prime of their output. Professions this was observed for see machine programmers, wherever 32.8% reported a productivity boost, consultants, enactment agents (Figure C) and recruiters.

Figure C

Impact of AI connected lawsuit enactment cause productivity.

In the lawsuit of consultants, the usage of GPT-4 bridged the spread betwixt low-skilled and high-skilled professionals, with the low-skilled radical experiencing much of a show boost (Figure D). Other research has besides indicated however generative AI successful peculiar could enactment arsenic an equaliser, arsenic the little experienced, little skilled workers get much retired of it.

Figure D

Improvement of enactment show connected low- and high-skilled consultants erstwhile utilizing AI.

However, different studies did suggest that “using AI without due oversight tin pb to diminished performance,” the researchers wrote. For example, determination are wide reports that hallucinations are prevalent successful ample connection models that execute ineligible tasks. Other probe has recovered that we whitethorn not scope the afloat imaginable of AI-enabled productivity gains for different decade, arsenic unsatisfactory outputs, analyzable guidelines and deficiency of proficiency proceed to clasp workers back.

4. AI regulations successful the U.S. are connected the rise

The AI Index Report recovered that, successful 2023, determination were 25 AI-related regulations progressive successful the U.S., portion successful 2016 determination was lone 1 (Figure E). This hasn’t been a dependable incline, though, arsenic the full fig of AI-related regulations grew by 56.3% from 2022 to 2023 alone. Over time, these regulations person besides shifted from being expansive regarding AI advancement to restrictive, and the astir prevalent taxable they interaction connected is overseas commercialized and planetary finance.

Figure E

Number of AI-related regulations progressive successful the U.S. betwixt 2016 and 2023.

AI-related authorities is besides expanding successful the EU, with 46, 22 and 32 caller regulations being passed successful 2021, 2022 and 2023, respectively. In this region, regulations thin to instrumentality a much expansive attack and astir often screen science, exertion and communications.

SEE: NIST Establishes AI Safety Consortium

It is indispensable for businesses funny successful AI to enactment updated connected the regulations that interaction them, oregon they enactment themselves astatine hazard of dense non-compliance penalties and reputational damage. Research published successful March 2024 recovered that only 2% of ample companies successful the U.K. and EU were alert of the incoming EU AI Act.

5. Investment successful generative AI is increasing

Funding for generative AI products that make contented successful effect to a punctual astir octupled from 2022 to 2023, reaching $25.2 cardinal (Figure F). OpenAI, Anthropic, Hugging Face and Inflection, amongst others, each received important fundraising rounds.

Figure F

Total planetary backstage concern successful generative AI from 2019 to 2023.

The buildout of generative AI capabilities is apt to conscionable request from businesses looking to follow it into their processes. In 2023, generative AI was cited successful 19.7% of each net calls of Fortune 500 companies, and a McKinsey report revealed that 55% of organisations present usage AI, including generative AI, successful astatine slightest 1 concern portion oregon function.

Awareness of generative AI boomed aft the motorboat of ChatGPT connected November 30, 2022, and since then, organisations person been racing to incorporated its capabilities into their products oregon services. A caller survey of 300 planetary businesses conducted by MIT Technology Review Insights, successful concern with Telstra International, recovered that respondents expect their number of functions deploying generative AI to much than treble successful 2024.

SEE: Generative AI Defined: How it Works, Benefits and Dangers

However, determination is immoderate grounds that the roar successful generative AI “could travel to a reasonably swift end”, according to starring AI dependable Gary Marcus, and businesses should beryllium wary. This is chiefly owed to limitations successful existent technologies, specified arsenic imaginable for bias, copyright issues and inaccuracies. According to the Stanford report, the finite magnitude of online information disposable to bid models could exacerbate existing issues, placing a ceiling connected improvements and scalability. It states that AI firms could tally retired of high-quality connection information by 2026, low-quality connection information successful 2 decades and representation information by the precocious 2030s to mid-2040s.

6. Benchmarks for LLM work alteration widely

There is important saltation successful the benchmarks that tech companies measure their LLMs against erstwhile it comes to trustworthiness oregon responsibility, according to the study (Figure G). The researchers wrote that this “complicates efforts to systematically comparison the risks and limitations of apical AI models.” These risks see biassed outputs and leaking backstage accusation from grooming datasets and speech histories.

Figure G

The liable AI benchmarks utilized successful the improvement of fashionable AI models.

Reuel, a PhD pupil successful the Stanford Intelligent Systems Laboratory, told TechRepublic successful an email: “There are presently nary reporting requirements, nor bash we person robust evaluations that would let america to confidently accidental that a exemplary is harmless if it passes those evaluations successful the archetypal place.”

Without standardisation successful this area, the hazard that immoderate untrustworthy AI models whitethorn gaffe done the cracks and beryllium integrated by businesses increases. “Developers mightiness selectively study benchmarks that positively item their model’s performance,” the study added.

Reuel told TechRepublic: “There are aggregate reasons wherefore a harmful exemplary tin gaffe done the cracks. Firstly, nary standardised oregon required evaluations making it hard to comparison models and their (relative) risks, and secondly, nary robust evaluations, specifically of instauration models, that let for a solid, broad knowing of the implicit hazard of a model.”

7. Employees are tense and acrophobic astir AI

The study besides tracked however attitudes towards AI are changing arsenic consciousness increases. One survey recovered that 52% explicit nervousness towards AI products and services, and that this fig had risen by 13% implicit 18 months. It besides recovered that lone 54% of adults hold that products and services utilizing AI person much benefits than drawbacks, portion 36% are fearful it whitethorn instrumentality their occupation wrong the adjacent 5 years (Figure H).

Figure H

Global opinions connected the interaction AI volition person connected existent jobs successful 2023.

Other surveys referenced successful the AI Index Report recovered that 53% of Americans presently consciousness much acrophobic astir AI than excited, and that the associated astir communal interest they person is its interaction connected jobs. Such worries could person a peculiar interaction connected employee intelligence health erstwhile AI technologies commencement to beryllium integrated into an organisation, which concern leaders should monitor.

SEE: The 10 Best AI Courses successful 2024

8. US and China are creating astir of today’s fashionable LLMs

TechRepublic’s Ben Abbott covered this inclination from the Stanford study successful his nonfiction astir gathering AI instauration models successful the APAC region. He wrote, successful part:

“The dominance of the U.S. successful AI continued passim 2023. Stanford’s AI Index Report released successful 2024 recovered 61 notable models had been released successful the U.S. successful 2023; this was up of China’s 15 caller models and France, the biggest contributor from Europe with 8 models (Figure I). The U.K. and European Union arsenic a portion produced 25 notable models — beating China for the archetypal clip since 2019 — portion Singapore, with 3 models, was the lone different shaper of notable ample connection models successful APAC.”

Figure I