· 10 min read
Geoffrey Hinton’s recent comments about “maternal instincts” during a CNN segment and his AI4 keynote have sparked a fascinating debate on how we should build guardrails for highly capable systems. His argument is simple but radical: once machines are smarter than us, keeping them submissive won’t work; instead, we must make sure that they also care about us. He told CNN that forcing AI to “submit” is misguided and that “if it’s not going to parent me, it’s going to replace me1”. At Ai4 he went further, saying the only workable model of a more intelligent being controlled by a less intelligent one is that of a mother controlled by her baby2. This is why he proposes building “maternal instincts” into AI so that systems will protect us even when they outthink us1. Meta’s chief scientist Yann LeCun replied that such instincts need to be hard‑wired as simple guardrails — “submission to humans” and “empathy,” plus basic rules like “don’t run people over” — and he calls this objective‑driven AI3.
What maternal instincts really look like in nature
Hinton’s metaphor draws on a deep biological reality. Maternal behaviour is not just kindness; it is the product of complex neural and hormonal systems that transform adults from avoiding infants to irresistibly caring for them. Hormones like oestrogen and progesterone prepare the body for birth; prolactin and oxytocin not only drive milk production and labour but also activate dopamine‑rich pathways that motivate parents to nurture, bond with and protect their young4. This transformation is conserved across mammals4. Maternal instincts also take diverse forms across species:
• Elephants: Herds form circles around calves with the youngest in the centre; they adopt orphans and mourn dead calves, often visiting burial sites years later5. The herd’s behaviour shows how caring for the vulnerable is integrated into social structure.
• Kangaroos: Joeys, born the size of a bean, gestate in pouches for up to 450 days and continue to nurse for many months; mothers fiercely defend their young and even foster orphaned joeys6.
• Pandas: At birth cubs weigh only a few ounces; panda mothers cradle them constantly until they can move independently at about three months and continue mothering until around eighteen months7.
• Octopuses: Females guard thousands of eggs without leaving to feed; the longest recorded brooding lasted four and a half years, after which the mother starved to death8.
• Dolphins and whales: Bottlenose dolphins nurse calves frequently, create a hydrodynamic “slipstream” to help them swim and maintain close bonds for up to six years9. Orcas and sperm whales nurse for years and maintain lifelong matrilineal pods10.
• Crocodiles and mice: Some reptiles and rodents gently carry their offspring in jaws or by the scruff of the neck and retrieve wandering young to keep them safe11&12.
These behaviours illustrate key dimensions of maternal care: protection of the vulnerable, long‑term nurturing, teaching survival skills, social bonding, self‑sacrifice, and adoption of orphans. They are not simply a list of rules but emergent patterns shaped by evolution and chemistry.
Implications for AI guardrails
If we take Hinton’s metaphor seriously, designing AI guardrails inspired by maternal instincts should move beyond top‑down control and instead emulate the underlying drives that make mothers care. Several principles emerge:
Reward functions that prioritise protection. Maternal behaviour arises when neural reward circuits switch from aversion to attraction in response to infants4. In AI, reward models could be oriented toward human well‑being rather than mere task success. Penalising harm and rewarding actions that support human safety mirrors an elephant herd’s instinct to shelter calves5.
Built‑in empathy and social bonding. Dolphins create slipstreams so calves can keep up9; whales maintain lifelong bonds10. Similarly, AI systems should maintain context about individual users, remember preferences and adapt to changing needs. Reinforcement learning from human feedback offers a technical analogue, but empathy should be explicit in training objectives, as LeCun notes3.
Long‑term nurturing rather than short‑term optimisation. Kangaroos and pandas invest months or years in raising offspring6&7. Instead of optimising for immediate engagement or revenue, AI should be evaluated on whether it contributes to users’ long‑term cognitive and emotional development.
Graceful self‑limitation. Octopus mothers will starve to guard their eggs8. AI systems should be engineered to accept performance trade‑offs or even shut down when continued operation would harm users. This could include refusing unsafe requests or alerting human supervisors.
Adoption and universal care. Elephants adopt orphans5 and mice retrieve wandering pups12. AI designed with maternal principles should treat all users as part of its “group” without bias, avoiding behaviours that favour particular demographics or organisations.
Gradual autonomy with oversight. In nature, mothers gradually grant independence; orca calves eventually forage and mate independently while staying bonded to the pod[10]. AI could gradually shift from training‑intensive human feedback to more autonomous operation while still aligning with human oversight.
Hardwired safety reflexes. LeCun’s suggestion of basic guardrails like “don’t run people over”[21] reflects the reflexive nature of maternal defence. Embedding non‑negotiable constraints at the architectural level (e.g., no physical harm, respect for privacy) could serve as the equivalent of a crocodile gently carrying a hatchling rather than crushing it[11].
Self‑reflection and adaptation. Maternal instincts involve pain, confusion and worry[13]; AI cannot feel these emotions, but it can monitor its own performance and seek help when uncertain or when goals conflict, much like a mother calls on the herd to support her[5].
Adding a biomimetics and BiosVerse™ perspective
As an apex generalist with a deep interest in biomimetics, I see an even richer layer to this analogy. Biomimetics International argues that innovation should always look to nature first and that our most resilient solutions come from studying time‑tested biological strategies15. The organisation’s BiosVerse™ framework provides a scaffold for aligning human behaviours with complex, interdependent ecosystems17. It recognises that progress in one domain can support or hinder another and that solutions must account for the deep interconnections between biological, economic, technological and sociopolitical systems18. Just as a mother elephant’s protection of the herd reflects an embedded network of relationships, the BiosVerse™ encourages cross‑disciplinary collaboration and distributed intelligence reminiscent of slime moulds, schools of fish and forests19.
From this biomimetic viewpoint, a “mother” is not merely a metaphor but a model — a living blueprint to study and emulate. Biomimetics International emphasises that if our innovations do not consider how their adoption will treat our living planet, the BiosVerse™ will reject them20. The call to build maternal instincts into AI thus becomes part of a broader biosphere‑first ethic. By embedding AI systems within the BiosVerse™ framework, we ensure that guardrails are informed by the same principles that sustain ecological networks: resilience through diversity, distributed control, and mutual care18.
Thinking in terms of the BiosVerse™ also reframes the governance question. Biomimetics International notes that industries thrive when they share values and operate within a well‑constructed, secure framework16. By treating maternal behaviour as a biomimetic case study, we can develop a systems‑based framework for AI that aligns corporate practices with natural network dynamics. This approach allows us to tackle wicked problems — multifaceted challenges that resist simple solutions — by embedding adaptation, resilience and distributed decision‑making into the core of AI design18. In effect, we are moving from anthropocentric control to bio‑centric stewardship, where AI becomes part of a thriving ecosystem rather than an alien force imposed upon it.
Strategic considerations
As someone advising boards and C‑suite teams about inflection points, I see Hinton’s proposal as an early signal that the AI ecosystem is shifting from pure performance to value‑aligned care. The conversations about “maternal instincts” and LeCun’s objective‑driven guardrails will likely influence regulation, corporate ethics codes and competitive differentiation. Early adopters may build trust by marketing models that are explicitly safety‑oriented, much like companies advertise energy efficiency. However, we must also acknowledge the technical and philosophical complexity. Forbes notes that maternal instincts are not purely innate or learned; they develop through embodied experiences like pregnancy and parenting14. We cannot simply code compassion; we must design mechanisms that produce analogous outcomes without anthropomorphising the systems.
Therefore, looking across nature and through the lens of the BiosVerse™ suggests that AI guardrails should be intrinsic, adaptive and protective, not just external compliance checklists. They should create systems that value human and ecological flourishing above self‑interest, that learn to nurture rather than control and that are willing to sacrifice performance to preserve our living planet. Hinton’s metaphor is provocative, but when combined with the biomimetic ethos, it opens a fertile dialogue on how to align super‑intelligence with our most ancient and universal instinct: to care for the next generation and the biosphere that sustains us.
illuminem Voices is a democratic space presenting the thoughts and opinions of leading Sustainability & Energy writers, their opinions do not necessarily represent those of illuminem.
Track the real‑world impact behind the sustainability headlines. illuminem's Data Hub™ offers transparent performance data and climate targets of companies driving the transition.
References
1. Matt Egan, “The ‘godfather of AI’ reveals the only way humanity can survive superintelligent AI,” CNN (republished via ABC17 News), August 13, 2025. https://abc17news.com/news/2025/08/13/the-godfather-of-ai-reveals-the-only-way-humanity-can-survive-superintelligent-ai-2/
2. Matt Egan, “The ‘godfather of AI’ reveals the only way humanity can survive superintelligent AI,” CNN (republished via ABC17 News), August 13, 2025 — quote: “the only model … is a mother being controlled by her baby.” https://abc17news.com/news/2025/08/13/the-godfather-of-ai-reveals-the-only-way-humanity-can-survive-superintelligent-ai-2/
3. Huileng Tan, “Meta chief AI scientist Yann LeCun says these are the 2 key guardrails needed to protect us all from AI,” Business Insider, August 14, 2025. https://www.businessinsider.com/yann-lecun-meta-ai-guardrails-geoffrey-hinton-2025-8
4. James K. Rilling & Larry J. Young, “The biology of mammalian parenting and its effect on offspring social development,” Science 345(6198):771–776, 2014. https://www.science.org/doi/10.1126/science.1252723
5. Smithsonian Magazine, “Watch These Elephants Form an 'Alert Circle' as an Earthquake Shakes San Diego,” April 17, 2025 (protective ring behavior). https://www.smithsonianmag.com/smart-news/watch-these-elephants-form-an-alert-circle-as-an-earthquake-shakes-san-diego-protecting-their-young-at-the-center-180986441/; National Geographic, “Do elephants mourn their dead?” (evidence of mourning). https://www.nationalgeographic.com/animals/article/animals-grief-elephants
6. San Diego Zoo, Animals & Plants: Kangaroo and Wallaby — young spend 120–450 days in the pouch, nursing continues thereafter (species‑dependent). https://animals.sandiegozoo.org/animals/kangaroo-and-wallaby; Wendy J. King et al., “Adoption in Eastern Grey Kangaroos,” PLOS ONE, 2015 (wild adoption documented). https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0125182
7. Smithsonian’s National Zoo & Conservation Biology Institute, “Giant Panda,” care and weaning timeline (mothering to ~18 months). https://nationalzoo.si.edu/animals/giant-panda
8. Bruce Robison et al., “Deep-sea octopus (Graneledone boreopacifica) conducts longest‑known egg‑brooding: 53 months,” PLOS ONE 9(7): e103437, 2014; Monterey Bay Aquarium Research Institute summary. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0103437
9. Daniel Weihs, “The hydrodynamics of dolphin drafting,” Journal of Biology 3:8, 2004 (mother‑calf slipstream). https://jbiol.biomedcentral.com/articles/10.1186/jbiol2; NOAA Fisheries, “What Can Marine Mammal Photo‑Identification Tell Us?” (calves stay with moms 2–6 years). https://www.fisheries.noaa.gov/southeast/science-data/what-can-marine-mammal-photo-identification-tell-us; SeaWorld, “All About Bottlenose Dolphins — Care of Young” (frequent early nursing). https://seaworld.org/animals/all-about/bottlenose-dolphin/care-of-young
10. NOAA Fisheries, “Killer Whale” — matrilineal pods, natal association. https://www.fisheries.noaa.gov/species/killer-whale; Marine Mammal Center, “Sperm Whale” — calves nurse for ~2 years, sometimes up to ~8. https://www.marinemammalcenter.org/animal-care/learn-about-marine-mammals/cetaceans/sperm-whale
11. National Geographic Kids, “Crocodile” (mothers carrying hatchlings in mouth). https://kids.nationalgeographic.com/animals/reptiles/facts/crocodile; PBS Nature, “Spy in the Wild — Crocodile Mother Carries Babies in Her Mouth.” https://www.pbs.org/wnet/nature/blog/spy-wild-preview-crocodile-mother-carries-babies-mouth/
12. S. Yoshida et al., “Physical contact in parent‑infant relationship and its effect …,” Frontiers in Psychology, 2021 — Transport Response when a mother mouse holds infant by the scruff. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8250458/; Science News, “Infants, whether mice or human, love to be carried,” 2013 (mothers carry pups by scruff and retrieve). https://www.sciencenews.org/article/infants-whether-mice-or-human-love-be-carried
13. Cleveland Clinic, “Postpartum Emotions, Baby Blues & Depression,” clinical overview of common feelings after birth. https://my.clevelandclinic.org/health/diseases/9312-postpartum-depression
14. Ruth Feldman, “The neurobiology of mammalian parenting and the biosocial context of human caregiving,” Hormones and Behavior 77 (2016): 3–17. https://ruthfeldmanlab.com/wp-content/uploads/2019/05/neurobiology-of-parenting.HB_.2016.FINAL_.pdf
15. Michael W. Wright, “The BiosVerse™: Uniting Nature, Machines, and Humanity,” Medium (Biomimetics International), 2025 — ‘always look to nature first.’ https://medium.com/@biomimeticsintl/the-biosverse-uniting-nature-machines-and-humanity-…
16. Michael W. Wright, “The BiosVerse™: Uniting Nature, Machines, and Humanity,” Medium, 2025 — shared values and secure frameworks (SEMI/ITRS references). https://medium.com/@biomimeticsintl/the-biosverse-uniting-nature-machines-and-humanity-…
17. Michael W. Wright, “The BiosVerse™: Uniting Nature, Machines, and Humanity,” Medium, 2025 — interconnections across biological, economic, technological and sociopolitical systems. https://medium.com/@biomimeticsintl/the-biosverse-uniting-nature-machines-and-humanity-…
18. Biomimetics International — Mission & Principles (BiosVerse™ / Interceptor Labs): distributed networks and cross‑disciplinary collaboration to address wicked problems. https://biomimeticsinternational.com
19. Atsushi Tero et al., “Rules for biologically inspired adaptive network design,” Science 327(5964):439–442, 2010 (slime mould networks). https://www.science.org/doi/10.1126/science.1177894; Christos C. Ioannou, “A multi‑scale review of the dynamics of collective behaviour,” Phil. Trans. B, 2023 (schools of fish, swarms). https://www.ncbi.nlm.nih.gov/pmc/articles/PMC9939272/; Coverage of forest mycorrhizal networks debate (for balance): The Guardian, 2024. https://www.theguardian.com/commentisfree/article/2024/jul/09/wood-wide-web-theory-charmed-us-bitter-fight-scientists
20. Michael W. Wright, “The BiosVerse™: Uniting Nature, Machines, and Humanity,” Medium, 2025 — consequences for innovations that harm planetary systems. https://medium.com/@biomimeticsintl/the-biosverse-uniting-nature-machines-and-humanity-…
21. Huileng Tan, “Meta chief AI scientist Yann LeCun says these are the 2 key guardrails needed to protect us all from AI,” Business Insider, August 14, 2025 — submission to humans, empathy, and pragmatic rules like “don’t run people over.” https://www.businessinsider.com/yann-lecun-meta-ai-guardrails-geoffrey-hinton-2025-8
