Hujja don Hankali na Wucin Gadi na Psychometric

1. Teburin Abubuwan Ciki

• 2. Gabatarwa
• 3. Babban Fahimta: Canjin Tsarin Psychometric
• 4. Tsarin Tunani: Daga AI Mai Ƙuntata zuwa Hankali na Gabaɗaya
• 5. Ƙarfi da Rashi: Nazari Mai Mahimmanci na Gwajin AGI
• 6. Bayanai Masu Aiki: Hanyoyi na Gaba
• 7. Bayanan Fasaha da Tsarin Lissafi
• 8. Sakamakon Gwaji da Nazarin Ma'auni
• 9. Tsarin Nazari: Nazarin Misali na ARC
• 10. Aikace-aikace na Gaba da Hasashe
• 11. Nazari na Asali da Sharhi
• 12. Manazarta

2. Gabatarwa

Takardar "Hujja don Hankali na Wucin Gadi na Psychometric" ta Mark McPherson (Jami'ar Bournemouth, 2020) tana nazarin gwaje-gwaje da ma'auni na yanzu don auna Hankali na Wucin Gadi na Gabaɗaya (AGI). Mawallafin yana jayayya cewa tsarin AI na yanzu, duk da cewa suna samun aikin da ya fi na ɗan adam a fannoni masu ƙuntata kamar Go, StarCraft, da ganewar cuta, ba su da ikon daidaitawa da haɓakawa na hankalin ɗan adam. Babban ra'ayin shine cewa hanyoyin psychometric, musamman ma Abstraction and Reasoning Corpus (ARC) wanda Chollet ya gabatar, suna ba da hanya mafi kyau don gano da auna AGI.

3. Babban Fahimta: Canjin Tsarin Psychometric

Babban fahimtar wannan takarda ita ce cewa auna AGI yana buƙatar canjin tsari daga ma'auni na musamman na aiki zuwa tsarin psychometric wanda ke tantance iyawar fahimi na gabaɗaya. Mawallafin yana jayayya cewa ma'auni na AI na gargajiya (misali, wasa, rarraba hoto) ba su isa ba saboda suna auna aikin ƙuntata, na musamman na fanni maimakon hankali na gabaɗaya. Hanyar psychometric, wacce aka yi wahayi daga gwajin hankalin ɗan adam, tana mai da hankali kan auna ikon warware matsalolin sababbi a cikin fannoni daban-daban ba tare da horo na musamman na aiki ba.

4. Tsarin Tunani: Daga AI Mai Ƙuntata zuwa Hankali na Gabaɗaya

Takardar tana bin ci gaba mai ma'ana:

1. Gano Matsala: Tsarin AI na yanzu suna da ƙuntata kuma marasa ƙarfi, suna kasawa lokacin da yanayi ya ɗan bambanta daga yanayin horo.
2. Ma'anar AGI: Hankali na gabaɗaya an ayyana shi azaman ikon yin ayyuka a cikin fannoni da yawa, gami da waɗanda ba a san su a lokacin ƙirƙira ba.
3. Nazarin Gwaje-gwajen da Ake da su: Mawallafin yana kimanta gwaje-gwaje shida da Mikhaylovskiy ya gabatar (Bayani, Saita Matsala, Ƙaryata, Hasashen Sabon Al'amari, Ƙirƙirar Kasuwanci, Ƙirƙirar Ka'idar) da ma'aunin ARC na Chollet.
4. Nazari Mai Mahimmanci: Kowane gwaji ana kimanta shi bisa ga ma'auni ciki har da gama gari, rashin son rai, haɓakawa, da juriya ga yaudara.
5. Shawarwari: Hanyoyin psychometric, musamman ARC, an gano su a matsayin hanya mafi kyau.

5. Ƙarfi da Rashi: Nazari Mai Mahimmanci na Gwajin AGI

5.1 Ƙarfin Hanyoyin Psychometric

• Gama gari: Ayyukan ARC suna buƙatar tunani game da tsarin da ba a taɓa gani ba, ba ilimin musamman na fanni ba.
• Rashin son rai: Ana auna aiki ta hanyar nasara akan ayyukan da ba a gani ba, yana rage son rai.
• Haɓakawa: Tarin bayanan ARC yana ƙunshe da ayyuka 800, yana ba da damar nazarin ƙididdiga mai ƙarfi.

5.2 Rashi da Iyakoki

• Gwajin Mikhaylovskiy: Gwajin Bayani, Ƙirƙirar Ka'idar, da Ƙirƙirar Kasuwanci sun fi son ɗan adam da wuya a sarrafa su ta atomatik. Suna buƙatar ƙirƙira matakin ɗan adam da hulɗa da duniya ta gaske, wanda bazai zama dole ba don AGI.
• Iyakokin ARC: Duk da cewa yana da alƙawari, ARC yana mai da hankali ne kawai akan tunani na gani kuma bazai iya ɗaukar wasu nau'ikan hankali ba (misali, zamantakewa, harshe, ko tunani na jiki).
• Rashin Canjin Lokaci: Yawancin gwaje-gwaje suna tsaye ne kuma basa tantance koyo akan lokaci ko daidaitawa ga yanayi masu canzawa.

6. Bayanai Masu Aiki: Hanyoyi na Gaba

Dangane da binciken, takardar tana ba da shawarar hanyoyi masu aiki da yawa:

• Haɓaka Ma'auni Masu Haɗaka: Haɗa ayyukan psychometric tare da yanayi masu canzawa, masu mu'amala don tantance duka tunani da daidaitawa.
• Haɗa Hanyoyi da yawa: Fadada ARC don haɗa ayyukan tunani na harshe, sauti, da jiki.
• Mai da hankali kan Haɗaɗɗen Gabaɗaya: Tsara ayyukan da ke buƙatar haɗa abubuwan da aka koya ta hanyoyi sababbi, wani muhimmin al'amari na hankalin ɗan adam.
• Ɗauki Rahoton Daidaitacce: Yi amfani da ma'aunin psychometric (misali, dogaro, inganci, ka'idar amsa abu) don tabbatar da cewa ma'auni suna da tsauri a kimiyyance.

7. Bayanan Fasaha da Tsarin Lissafi

Hanyar psychometric don auna AGI za a iya tsara ta ta amfani da Ka'idar Amsa Abu (IRT). Bari $ heta$ ya wakilci hankali na gabaɗaya na wakili. Yiwuwar warware aiki $i$ da wahala $b_i$ da bambanci $a_i$ ana bayar da shi ta hanyar samfurin logistic:

$$P(X_i = 1 | \theta) = \frac{1}{1 + e^{-a_i(\theta - b_i)}}$$

Don ma'aunin ARC, kowane aiki ya ƙunshi nau'i-nau'i na grid shigarwa da fitarwa. Wakili dole ne ya gano canjin da ke ƙasa $f: \mathbb{Z}^{m \times n} \rightarrow \mathbb{Z}^{p \times q}$ daga wasu misalai kuma ya yi amfani da shi akan sabon shigarwa. Ma'aunin aiki shine daidaito akan ayyukan da aka ajiye, wanda aka auna shi da wahalar aiki.

8. Sakamakon Gwaji da Nazarin Ma'auni

Takardar ba ta gabatar da gwaje-gwaje na asali ba amma tana nazarin sakamakon da ake da su. Babban binciken daga wallafe-wallafen ya haɗa da:

• Aikin ɗan Adam akan ARC: Mutane suna samun kusan daidaito 80-90% akan ayyukan ARC, yana nuna yuwuwar ma'aunin.
• Aikin AI: Tsarin AI na zamani (tun daga 2020) suna samun ƙasa da daidaito 30% akan ARC, yana nuna tazara tsakanin hankali mai ƙuntata da na gabaɗaya.
• Kwatanta da Sauran Ma'auni: ARC ya fi ƙalubale fiye da gwajin IQ na gargajiya ga AI saboda yana buƙatar tunani irin na shirin maimakon daidaita tsari.

Hoto 1: Hoton ginshiƙi na hasashe yana kwatanta aikin ɗan adam da na AI akan ayyukan ARC a matakan wahala (sauƙi, matsakaici, wuya). Mutane suna ci gaba da yin fiye da AI, tare da faɗaɗa tazara akan ayyuka masu wuya.

9. Tsarin Nazari: Nazarin Misali na ARC

Don kwatanta hanyar psychometric, yi la'akari da aikin ARC inda shigarwa grid 3x3 ne mai sel masu launi, kuma fitarwa grid 3x3 ne mai wani tsari daban. Wakili dole ne ya gano doka (misali, "juya tsarin digiri 90 zuwa agogo") daga misalai biyu kuma ya yi amfani da shi akan shigarwa ta uku.

Misalin Aiki:

• Shigarwa 1: [[0,1,0],[1,0,1],[0,1,0]] → Fitarwa 1: [[0,1,0],[1,0,1],[0,1,0]] (babu canji, daidaito)
• Shigarwa 2: [[1,0,0],[0,1,0],[0,0,1]] → Fitarwa 2: [[0,0,1],[0,1,0],[1,0,0]] (juyawa tare da anti-diagonal)
• Gwajin Shigarwa: [[0,0,1],[0,1,0],[1,0,0]] → Fitarwa da ake tsammani: [[1,0,0],[0,1,0],[0,0,1]]

Wannan aikin yana buƙatar wakili ya gane dokar canji (juyawa tare da anti-diagonal) kuma ya yi amfani da ita akan sabon tsari. Ƙimar psychometric tana cikin gaskiyar cewa doka ba ta da tushe kuma ba ta da alaƙa da kowane fanni na musamman.

10. Aikace-aikace na Gaba da Hasashe

Hanyar psychometric zuwa AGI tana da aikace-aikace masu ban sha'awa da yawa:

• Tsaron AI: Ma'auni na psychometric na iya taimakawa gano gazawar da ba a zata ba a cikin tsarin AI ta hanyar gwada haɓakawa zuwa yanayin sababbi.
• Haɗin gwiwar ɗan Adam da AI: Fahimtar bayanin fahimi na AI (misali, ƙarfi a cikin tunani na gani vs. harshe) na iya inganta haɗin gwiwa tare da mutane.
• AI na Ilimi: Tsarin psychometric na iya jagorantar haɓaka malaman AI waɗanda ke daidaita da salon koyo na mutum ɗaya.
• Kimiyyar Jijiya: Kwatanta aikin ɗan adam da na AI akan ayyukan psychometric na iya ba da haske kan tushen jijiya na hankali na gabaɗaya.

Hanyoyi na gaba sun haɗa da haɗa ma'auni na psychometric tare da yanayin koyo na ƙarfafawa, haɓaka gwaje-gwaje masu canzawa waɗanda ke daidaita da matakin iyawar wakili, da ƙirƙirar ma'auni masu yawa waɗanda ke tantance tunani a cikin hanyoyin ji.

11. Nazari na Asali da Sharhi

Takardar tana ba da hujja mai ƙarfi don hanyoyin psychometric zuwa AGI, amma batutuwa masu mahimmanci da yawa suna buƙatar bincike. Na farko, dogaro da hankali irin na ɗan adam a matsayin ma'aunin zinariya abu ne mai shakka a falsafa. Kamar yadda Bostrom (2014) ya yi jayayya a cikin "Superintelligence," AGI na iya nuna nau'ikan hankali waɗanda suka bambanta da fahimtar ɗan adam, yana sa ma'auni na son ɗan adam ya zama mai yaudara. Na biyu, ma'aunin ARC, duk da kyawunsa, na iya zama mai ƙuntata sosai. Kamar yadda Lake et al. (2017) suka lura a cikin "Building Machines That Learn and Think Like People," hankalin ɗan adam ya ƙunshi ba kawai tunani na abstract ba har ma da ilimin lissafi na zahiri, fahimtar zamantakewa, da fahimtar harshe. Ma'aunin hankali na gabaɗaya na gaske ya kamata ya ƙunshi waɗannan bangarorin. Na uku, takardar ta yi watsi da yuwuwar gwajin adawa. Kamar yadda Goodfellow et al. (2014) suka nuna a cikin takardar GAN ta asali, misalan adawa na iya bayyana raunin asali a cikin tsarin AI wanda ma'auni na yau da kullun ya rasa. Haɗa abubuwan adawa cikin gwaje-gwajen psychometric na iya samar da ingantaccen kimantawa na haɓakawa. A ƙarshe, mayar da hankalin takardar kan auna maimakon gine-gine yana da ƙarfi, amma yana da haɗarin yin watsi da tambayar yadda ake gina AGI. Kamar yadda Yudkowsky (2008) ya yi jayayya, matsalar daidaitawa tana buƙatar fahimtar hanyoyin ciki na tsarin AI, ba kawai halayen su na waje ba. Duk da waɗannan iyakokin, takardar tana ba da tsari mai mahimmanci don tunani game da kimantawar AGI kuma ta jaddada buƙatar ma'auni masu tsauri, ingantattun psychometric.

12. Manazarta

1. McCarthy, J., et al. (1956). A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence.
2. Silver, D., et al. (2016). Mastering the game of Go with deep neural networks and tree search. Nature, 529(7587), 484-489.
3. Vinyals, O., et al. (2019). Grandmaster level in StarCraft II using multi-agent reinforcement learning. Nature, 575(7782), 350-354.
4. Krizhevsky, A., et al. (2012). ImageNet classification with deep convolutional neural networks. NeurIPS.
5. Vaswani, A., et al. (2017). Attention is all you need. NeurIPS.
6. Esteva, A., et al. (2017). Dermatologist-level classification of skin cancer with deep neural networks. Nature, 542(7639), 115-118.
7. Marcus, G. (2018). Deep learning: A critical appraisal. arXiv:1801.00631.
8. Searle, J. (1980). Minds, brains, and programs. Behavioral and Brain Sciences, 3(3), 417-424.
9. Thomson, W. (1889). Popular Lectures and Addresses.
10. Adams, S., et al. (2012). Mapping the landscape of human-level artificial general intelligence. AI Magazine, 33(1), 25-42.
11. Goertzel, B. (2014). Artificial general intelligence: Concept, state of the art, and future prospects. Journal of Artificial General Intelligence, 5(1), 1-48.
12. Bringsjord, S., & Schimanski, B. (2003). What is artificial intelligence? Psychometric AI as an answer. IJCAI.
13. Mikhaylovskiy, N. (2020). Six tests for artificial general intelligence. arXiv:2005.05718.
14. Chollet, F. (2019). On the measure of intelligence. arXiv:1911.01547.
15. Bostrom, N. (2014). Superintelligence: Paths, Dangers, Strategies. Oxford University Press.
16. Lake, B. M., et al. (2017). Building machines that learn and think like people. Behavioral and Brain Sciences, 40, e253.
17. Goodfellow, I., et al. (2014). Generative adversarial nets. NeurIPS.
18. Yudkowsky, E. (2008). Artificial intelligence as a positive and negative factor in global risk. In Global Catastrophic Risks, Oxford University Press.