The biggest obstacle to artificial intelligence is the hype

Technology companies have been evaluated for decades based on a promise of exponential growth rather than actual earnings. That exponential growth has sputtered, so for these companies to maintain the price of their stock, they need to try to find new avenues of growth. Before the current AI era, there was virtual reality and before that there was blockchains (Vainio 2022). Both made huge promises they couldn’t meet, which lead to investors seeking opportunities elsewhere.

The difference between virtual reality and blockchains is that blockchains turned out not to have any meaningful potential (Flick 2022), while virtual reality does hold potential, but a rush to invest in it followed by a failure to find a market of proportional size meant that the natural growth of the field has been hindered. However, it is possible that the bubble did leave behind technology that can be utilized in the future.

AI seems to be in a similar situation. It has promise, but that promise is being hyped to absurd levels. The technologies and the mathematical principles used in large language models, such as ChatGPT, has been around for decades, but implementing on such scale has been impossible due to prohibitive costs. However, ChatGPT has been unable to mend this problem. Each query costs an estimated 36 cents, which is less than 5% of Google revenue for each search on their search engine (Patel & Ahmad 2023). The situation seems dire.

At the same time, many influential people have invested huge amounts of money into these technologies. As they have a platform, they can evangelize for the technology, while actual experts in the field can find it difficult to gain any sort of mainstream visibility.

Forget general AI for now

There is a market for AI, but at this moment it’s in narrow AI rather than general AI. The current systems are nowhere near human intelligence, and we should not try to fool ourselves into believing they are. As ChatGPT is essentially a word calculator, no matter how smart it seems, it is also solidly in the narrow AI camp despite claims that we are on the cusp of reaching general AI.

At the same time, this raises the question: Why should we push into general AI, when there is so much to explore with narrow AI? The problem with overpromising regarding the capabilities of AI systems is that they are being used in contexts where they can’t actually do the work (for example Buchanan 2023).

PATHFINDER is project at LAB in collaboration with University of Dubrovnik and formatio Private School in Liechtenstein. The goal of the project is to enhance the comprehension of AI technology in higher education by providing educators and students with the necessary information and resources for its efficient and ethical use. An important part of this is to understand the inherent limitations of the technology and to see past the marketing.

References

Buchanan, T. 2023. Dispatch pauses AI sports writing program. Axios Columbus. Cited 31 Jan 2024. Available at https://www.axios.com/local/columbus/2023/08/28/dispatch-gannett-ai-newsroom-tool

Flick, C. 2022. A critical professional ethical analysis of Non-Fungible Tokens (NFTs). Journal of Responsible Technology. Vol. 12, 100054. Cited 31 Jan 2024. Available at https://doi.org/10.1016/j.jrt.2022.100054

Patel, D. & Ahmad, A. 2023. The Inference Cost Of Search Disruption – Large Language Model Cost Analysis. SemiAnalysis. Cited 31 Jan 2024. Available at https://www.semianalysis.com/p/the-inference-cost-of-search-disruption

Vainio, A. 2022. Web3 on osoittautunut kuplaksi – jatkaako Metaverse trendiä? LAB Pro. Cited 31 Jan  2024. Available at https://www.labopen.fi/lab-pro/web3-on-osoittautunut-kuplaksi-jatkaako-metaverse-trendia/

viarami. 2023. Winkler, M. Chat gpr, scrabble, tiles. Pixabay. Cited 5 Feb 2024. Available at https://pixabay.com/photos/chat-gpt-scrabble-tiles-typography-8282253/