By
Stuart Kerr, Technology Correspondent, LiveAIWire

Ninety-five percent of the ocean floor has never been observed by
human eyes. The deep sea is the largest habitat on Earth and among the least
understood: cold, dark, under pressures that destroy conventional equipment,
and located at depths that make sustained human presence impossible. For most
of the history of oceanography, this inaccessibility was simply accepted as a
limit on what could be known. AI and autonomous underwater vehicles are
beginning to change that relationship between science and the
abyss.

The transformation of deep-sea research by intelligent systems is
less visible than AI’s impact on more accessible domains, but it is
methodologically significant. It is producing discoveries that are reshaping
understanding of ocean biology, geology, and chemistry, and it is doing so at
a pace and scale that the previous generation of ocean scientists would not
have considered possible.

Autonomous Underwater Vehicles and Machine
Perception

Autonomous underwater vehicles equipped with AI navigation,
obstacle avoidance, and mission management systems can survey deep-sea
environments at depths, durations, and spatial coverages that remotely
operated vehicles, tethered to surface ships and controlled by human pilots,
cannot match. The removal of the tether removes both the operational
constraint and the surface ship dependency that make conventional ROV
operations expensive and weather-dependent.

AI navigation systems in deep-sea AUVs must operate under
conditions that make most terrestrial autonomous navigation approaches
inapplicable. GPS does not penetrate seawater; acoustic positioning provides
only coarse localisation; the environment is three-dimensional, dynamic, and
contains objects — organisms, geological features, equipment from previous
expeditions — that may not appear in any training dataset. The navigation
systems developed for deep-sea autonomous operation have produced advances in
AI localisation and path planning under uncertainty that have applications
beyond oceanography.

Computer vision systems trained on underwater imagery now classify
seafloor habitats, identify species, detect geological features, and flag
anomalies for scientist review with accuracy that approaches expert human
performance on established species and feature categories. The bottleneck
that limited the volume of useful information produced by deep-sea camera
systems — the availability of trained researchers to review hours of footage
— is easing as AI classification tools reduce the review burden while
increasing throughput.

Species Discovery at Unprecedented Scale

Deep-sea species discovery has historically been constrained by
the rarity and cost of expeditions. AI analysis of accumulated imagery
archives from previous expeditions is now producing species identifications
and range extensions from footage that was collected years ago but never
fully analysed. Research published in journals including Frontiers
in Marine Science has described AI-enabled analyses of archival
deep-sea footage that identified previously uncatalogued species and
behaviour patterns from video collected over multiple decades of prior
expedition activity.

New expeditions deploying AI-equipped AUVs are generating
discovery rates that substantially exceed those of conventionally operated
surveys. The Ocean Census, a global initiative to discover deep-sea species,
is using AI image classification as a core component of its methodology,
aiming to describe thousands of new species from deep-water habitats within a
decade. The scale of that ambition would not be achievable without AI
analysis of the imagery volume that modern AUVs can generate.

What this means for conservation: species cannot be protected if
their existence is not known. The deep sea contains a significant proportion
of Earth’s biodiversity, and that biodiversity is under increasing pressure
from deep-sea mining interest, bottom trawling, and climate-driven changes in
ocean chemistry and temperature. Rapid species discovery enabled by AI
provides the biological baseline required for evidence-based conservation of
habitats that have previously been managed — or mismanaged — in ignorance
of what they contain.

Climate Monitoring and Ocean Carbon Science

The ocean absorbs roughly a quarter of the carbon dioxide emitted
by human activity, making it the largest operational carbon sink in the Earth
system. Understanding the mechanisms and limits of that carbon absorption is
critical for climate modelling and for assessing the feasibility of carbon
removal strategies. Deep-sea monitoring of carbon flux, ocean chemistry, and
biological productivity is a research priority whose data requirements exceed
what conventional ship-based sampling can provide.

AI-equipped autonomous platforms — gliders, profiling floats, and
seabed landers — are generating continuous monitoring data from ocean
systems at depths and over durations that would be impossible with
conventional instrumentation. The Argo float network, which deploys thousands
of autonomous profiling floats globally, generates data that AI analysis is
increasingly used to interpret and integrate with satellite and model data
into comprehensive ocean state assessments.

Research from the Woods
Hole Oceanographic Institution on ocean carbon cycling has
highlighted the role of deep-sea biological communities in carbon
sequestration, a role that is only beginning to be quantified at the spatial
and temporal resolution that management decisions require. AI analysis of
biological and chemical monitoring data is accelerating the development of
that quantification.

Deep-Sea Mining and the AI Dilemma

The same AI tools that are enabling discovery of deep-sea
ecosystems are also enabling the commercial exploitation of them. Deep-sea
polymetallic nodule fields, which contain minerals critical for battery
technology and renewable energy infrastructure, are being surveyed using
AI-equipped autonomous vehicles operated by mining companies. The detailed
seabed mapping that AI makes possible is simultaneously a scientific resource
and a prospecting tool.

The governance of deep-sea mining is contested internationally.
The International Seabed Authority, established by the UN Convention on the
Law of the Sea, is responsible for regulating mining in international waters
but has faced criticism from environmental organisations and some member
states for moving toward mining regulation before adequate environmental
baselines have been established. AI-enabled discovery of deep-sea
biodiversity — species that would be destroyed by mining activities — is
directly relevant to the governance debate, providing the biological
knowledge required to make evidence-based decisions about where and whether
mining should be permitted.

The tension between the economic interests that motivate deep-sea
resource extraction and the scientific and conservation interests that
motivate deep-sea biodiversity research is not resolved by AI capability; it
is sharpened by it. Better knowledge of what the deep sea contains makes the
consequences of destroying it more legible without automatically determining
how the competing interests at stake should be weighted. The same dynamic
operates in other AI-informed conservation contexts, from wildlife
surveillance to habitat monitoring: AI provides better information,
and better information raises the stakes of the decisions made on its
basis.

The Future of Human-AI Ocean Exploration

The long-term vision for AI in ocean science is a persistent
monitoring network: a combination of satellite observation, autonomous
surface vehicles, profiling floats, and seabed landers that together provide
continuous, global coverage of the ocean system at a resolution and temporal
density that would transform both scientific understanding and the capacity
to detect change. The data volumes that such a network would generate are
beyond the capacity of human analysis alone; AI is not an optional addition
to this vision but a prerequisite for it.

Human oceanographers will remain essential to this vision, not as
observers but as scientists: formulating hypotheses, designing experiments,
interpreting anomalies, and translating AI-generated findings into the
contextual understanding that transforms data into knowledge. The explorer
archetype — the scientist descending into the unknown in a submersible — is
being supplemented rather than replaced by the data scientist who navigates
petabytes of AI-analysed ocean imagery to find the needle of discovery in a
haystack of seafloor footage. The abyss is not becoming less mysterious as AI
illuminates more of it; it is, if anything, revealing how much more there is
to know. The broader pattern of AI
extending human sensing into environments where human presence is
impossible finds its most literal expression in the deep ocean,
where the technology is not augmenting human capability but substituting for
it in an environment where unaugmented human presence is simply not
possible.

The international governance of deep-sea AI research and resource
exploitation is developing through several parallel tracks. The High Seas
Treaty, agreed in 2023, establishes a framework for the conservation and
sustainable use of marine biodiversity in international waters that applies
to areas beyond national jurisdiction — the same areas where deep-sea mining
interest is most intense. AI-enabled biological baseline surveys are directly
relevant to the treaty’s environmental impact assessment requirements,
providing the species and habitat data against which mining applications can
be evaluated. The treaty’s implementation will determine whether AI-enabled
discovery of deep-sea biodiversity serves primarily to enable better-informed
exploitation decisions or to strengthen the conservation case against
activities that would destroy ecosystems before they are understood. The
tension between AI as a tool for environmental insight and AI as an enabler
of environmentally damaging economic activity is nowhere more
starkly illustrated than in the deep sea, where the same technology
simultaneously reveals what is there and enables the commercial interest in
extracting it.

About the Author

Stuart
Kerr is a technology correspondent at LiveAIWire, covering artificial
intelligence, emerging technologies, and their impact on society and
industry.