In June 2022, Google engineer Blake Lemoine made headlines by claiming that the company's LaMDA chatbot had achieved perception. Lemoine stated that the software had the conversational abilities of a precocious seven-year-old and that we should assume it has a similar understanding of the world.
LaMDA, later released to the public under the name Bard, is powered by a Large Language Model (LLM), which also serves as the engine for OpenAI's ChatGPT bot. Other major tech companies are also racing to deploy similar technologies.
While billions of people now have access to LLM, few seem to believe that they are conscious. Instead, as linguist and data scientist Emily Bender poetically phrases it, they are "random parrots" that convincingly chatter without comprehension. But what about the next generation of artificial intelligence (AI) systems and beyond?
Our team of philosophers, neuroscientists, and computer scientists focuses on the scientific theories of how current human consciousness works and has compiled a list of fundamental computational characteristics. From our perspective, no current system comes close to the threshold of consciousness, but at the same time, there is no clear evidence that future systems will not genuinely be conscious.
Indicators of Consciousness
Since computer pioneer Alan Turing proposed the "imitation game" in 1950, the ability to successfully mimic human conversation has often been seen as a sign of consciousness. This is typically because the task appears too difficult to accomplish without consciousness.
However, just as the chess computer Deep Blue defeated grandmaster Gary Kasparov in 1997, the fluent conversations of LLM may have merely shifted the goalposts. Is there a principled way to address the question of artificial intelligence consciousness that does not rely on our intuitions about the difficulties or specialness of human cognition?
Our recent white paper aims to tackle this question. We compare current scientific theories of human consciousness, compile a list of "indicator attributes," and then apply them to artificial intelligence systems.
We do not claim that systems with indicator attributes are necessarily conscious, but the more indicators present, the more seriously we should consider the claim of artificial intelligence consciousness.
The Computational Processes Behind Consciousness
What indicators are we looking for? We avoid overt behavioral criteria, such as the ability to converse with humans, as these criteria tend to be anthropocentric and easily faked.
Instead, we examine theories of computational processes that support human consciousness. These can tell us what types of information processing are required to support subjective experience.
For example, the "global workspace theory" posits that consciousness arises from a bottleneck with limited capacity that gathers information from various brain regions and selects and disseminates globally available information. The "recursive processing theory" emphasizes the feedback effects of subsequent processing on preceding processing.
Each theory proposes more specific indicators. Our final list includes 14 indicators, each focusing on a different aspect of how systems work rather than their behavior.
No Reason to Believe Current Systems Are Conscious
Our analysis suggests that there is no reason to believe that current artificial intelligence systems are conscious.
Some do meet some indicators. Systems using the transformer architecture (such as ChatGPT and similar tools) satisfy three indicators of the "global workspace theory" but lack the crucial ability for global dissemination. They also fail to meet most other indicators.
So, despite ChatGPT's impressive conversational abilities, there may be no consciousness present internally. Other architectures, at most, meet only a few criteria.
Most current architectures satisfy only a few indicators. However, for most indicators, at least one current architecture meets the requirements.
This suggests that there are no apparent technical barriers in principle to building artificial intelligence systems that meet most or all indicators.
The emergence of conscious artificial intelligence systems may be just a matter of time. Of course, many questions will arise once conscious systems appear.
Alphabet Q1 paid subscribers reached 350M, fueled by YouTube & Google One. YouTube ad revenue slightly missed at $9.88B amid Premium growth. Total Q1 revenue hit $109.9B, beating forecasts.
Anthropic releases Claude connectors for Adobe, Blender, and Ableton, enabling AI workflow automation and app control. It also sponsors the Blender Development Foundation.
Apple introduces an App Store model splitting 12-month plans into monthly installments. On iOS 26.4+, it lowers upfront costs without reducing total price, adding billing tracking.
Google launches Aadhaar digital IDs in Wallet for India, enabling secure QR/NFC verification. A phased partner rollout targets key sectors, while passport-based credentials expand globally.
YouTube tests "Ask YouTube," an AI tool blending text and video for complex queries. Initially for US Premium users, it supports interactive follow-ups, expanding Google's AI search.
Google grants US DoD classified AI access, joining OpenAI and xAI in defense deals amid Anthropic's dispute. Contractual safeguards against surveillance and lethal weapons remain unenforceable.
Mistral AI launches Workflows, a Temporal-powered orchestration platform for secure, Python-driven enterprise AI automation featuring human-in-the-loop approvals.